diff options
author | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-05-05 17:47:29 +0000 |
---|---|---|
committer | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-05-05 17:47:29 +0000 |
commit | 4f5791ebd03eaec1c7da0865a383175b05102712 (patch) | |
tree | 8ce7b00f7a76baa386372422adebbe64510812d4 /source3/locking | |
parent | Initial commit. (diff) | |
download | samba-4f5791ebd03eaec1c7da0865a383175b05102712.tar.xz samba-4f5791ebd03eaec1c7da0865a383175b05102712.zip |
Adding upstream version 2:4.17.12+dfsg.upstream/2%4.17.12+dfsgupstream
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'source3/locking')
-rw-r--r-- | source3/locking/brlock.c | 2019 | ||||
-rw-r--r-- | source3/locking/leases_db.c | 726 | ||||
-rw-r--r-- | source3/locking/leases_db.h | 80 | ||||
-rw-r--r-- | source3/locking/leases_util.c | 77 | ||||
-rw-r--r-- | source3/locking/locking.c | 1186 | ||||
-rw-r--r-- | source3/locking/posix.c | 1374 | ||||
-rw-r--r-- | source3/locking/proto.h | 215 | ||||
-rw-r--r-- | source3/locking/share_mode_lock.c | 2517 | ||||
-rw-r--r-- | source3/locking/share_mode_lock.h | 140 | ||||
-rw-r--r-- | source3/locking/share_mode_lock_private.h | 24 |
10 files changed, 8358 insertions, 0 deletions
diff --git a/source3/locking/brlock.c b/source3/locking/brlock.c new file mode 100644 index 0000000..68ade6b --- /dev/null +++ b/source3/locking/brlock.c @@ -0,0 +1,2019 @@ +/* + Unix SMB/CIFS implementation. + byte range locking code + Updated to handle range splits/merges. + + Copyright (C) Andrew Tridgell 1992-2000 + Copyright (C) Jeremy Allison 1992-2000 + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. +*/ + +/* This module implements a tdb based byte range locking service, + replacing the fcntl() based byte range locking previously + used. This allows us to provide the same semantics as NT */ + +#include "includes.h" +#include "system/filesys.h" +#include "lib/util/server_id.h" +#include "locking/proto.h" +#include "smbd/globals.h" +#include "dbwrap/dbwrap.h" +#include "dbwrap/dbwrap_open.h" +#include "serverid.h" +#include "messages.h" +#include "util_tdb.h" + +#undef DBGC_CLASS +#define DBGC_CLASS DBGC_LOCKING + +#define ZERO_ZERO 0 + +/* The open brlock.tdb database. */ + +static struct db_context *brlock_db; + +struct byte_range_lock { + struct files_struct *fsp; + TALLOC_CTX *req_mem_ctx; + const struct GUID *req_guid; + unsigned int num_locks; + bool modified; + struct lock_struct *lock_data; + struct db_record *record; +}; + +/**************************************************************************** + Debug info at level 10 for lock struct. +****************************************************************************/ + +static void print_lock_struct(unsigned int i, const struct lock_struct *pls) +{ + struct server_id_buf tmp; + + DBG_DEBUG("[%u]: smblctx = %"PRIu64", tid = %"PRIu32", pid = %s, " + "start = %"PRIu64", size = %"PRIu64", fnum = %"PRIu64", " + "%s %s\n", + i, + pls->context.smblctx, + pls->context.tid, + server_id_str_buf(pls->context.pid, &tmp), + pls->start, + pls->size, + pls->fnum, + lock_type_name(pls->lock_type), + lock_flav_name(pls->lock_flav)); +} + +unsigned int brl_num_locks(const struct byte_range_lock *brl) +{ + return brl->num_locks; +} + +struct files_struct *brl_fsp(struct byte_range_lock *brl) +{ + return brl->fsp; +} + +TALLOC_CTX *brl_req_mem_ctx(const struct byte_range_lock *brl) +{ + if (brl->req_mem_ctx == NULL) { + return talloc_get_type_abort(brl, struct byte_range_lock); + } + + return brl->req_mem_ctx; +} + +const struct GUID *brl_req_guid(const struct byte_range_lock *brl) +{ + if (brl->req_guid == NULL) { + static const struct GUID brl_zero_req_guid; + return &brl_zero_req_guid; + } + + return brl->req_guid; +} + +/**************************************************************************** + See if two locking contexts are equal. +****************************************************************************/ + +static bool brl_same_context(const struct lock_context *ctx1, + const struct lock_context *ctx2) +{ + return (server_id_equal(&ctx1->pid, &ctx2->pid) && + (ctx1->smblctx == ctx2->smblctx) && + (ctx1->tid == ctx2->tid)); +} + +bool byte_range_valid(uint64_t ofs, uint64_t len) +{ + uint64_t max_len = UINT64_MAX - ofs; + uint64_t effective_len; + + /* + * [MS-FSA] specifies this: + * + * If (((FileOffset + Length - 1) < FileOffset) && Length != 0) { + * return STATUS_INVALID_LOCK_RANGE + * } + * + * We avoid integer wrapping and calculate + * max and effective len instead. + */ + + if (len == 0) { + return true; + } + + effective_len = len - 1; + if (effective_len <= max_len) { + return true; + } + + return false; +} + +bool byte_range_overlap(uint64_t ofs1, + uint64_t len1, + uint64_t ofs2, + uint64_t len2) +{ + uint64_t last1; + uint64_t last2; + bool valid; + + /* + * This is based on [MS-FSA] 2.1.4.10 + * Algorithm for Determining If a Range Access + * Conflicts with Byte-Range Locks + */ + + /* + * The {0, 0} range doesn't conflict with any byte-range lock + */ + if (ofs1 == 0 && len1 == 0) { + return false; + } + if (ofs2 == 0 && len2 == 0) { + return false; + } + + /* + * The caller should have checked that the ranges are + * valid. But currently we gracefully handle + * the overflow of a read/write check. + */ + valid = byte_range_valid(ofs1, len1); + if (valid) { + last1 = ofs1 + len1 - 1; + } else { + last1 = UINT64_MAX; + } + valid = byte_range_valid(ofs2, len2); + if (valid) { + last2 = ofs2 + len2 - 1; + } else { + last2 = UINT64_MAX; + } + + /* + * If one range starts after the last + * byte of the other range there's + * no conflict. + */ + if (ofs1 > last2) { + return false; + } + if (ofs2 > last1) { + return false; + } + + return true; +} + +/**************************************************************************** + See if lck1 and lck2 overlap. +****************************************************************************/ + +static bool brl_overlap(const struct lock_struct *lck1, + const struct lock_struct *lck2) +{ + return byte_range_overlap(lck1->start, + lck1->size, + lck2->start, + lck2->size); +} + +/**************************************************************************** + See if lock2 can be added when lock1 is in place. +****************************************************************************/ + +static bool brl_conflict(const struct lock_struct *lck1, + const struct lock_struct *lck2) +{ + /* Read locks never conflict. */ + if (lck1->lock_type == READ_LOCK && lck2->lock_type == READ_LOCK) { + return False; + } + + /* A READ lock can stack on top of a WRITE lock if they have the same + * context & fnum. */ + if (lck1->lock_type == WRITE_LOCK && lck2->lock_type == READ_LOCK && + brl_same_context(&lck1->context, &lck2->context) && + lck1->fnum == lck2->fnum) { + return False; + } + + return brl_overlap(lck1, lck2); +} + +/**************************************************************************** + See if lock2 can be added when lock1 is in place - when both locks are POSIX + flavour. POSIX locks ignore fnum - they only care about dev/ino which we + know already match. +****************************************************************************/ + +static bool brl_conflict_posix(const struct lock_struct *lck1, + const struct lock_struct *lck2) +{ +#if defined(DEVELOPER) + SMB_ASSERT(lck1->lock_flav == POSIX_LOCK); + SMB_ASSERT(lck2->lock_flav == POSIX_LOCK); +#endif + + /* Read locks never conflict. */ + if (lck1->lock_type == READ_LOCK && lck2->lock_type == READ_LOCK) { + return False; + } + + /* Locks on the same context don't conflict. Ignore fnum. */ + if (brl_same_context(&lck1->context, &lck2->context)) { + return False; + } + + /* One is read, the other write, or the context is different, + do they overlap ? */ + return brl_overlap(lck1, lck2); +} + +#if ZERO_ZERO +static bool brl_conflict1(const struct lock_struct *lck1, + const struct lock_struct *lck2) +{ + if (lck1->lock_type == READ_LOCK && lck2->lock_type == READ_LOCK) { + return False; + } + + if (brl_same_context(&lck1->context, &lck2->context) && + lck2->lock_type == READ_LOCK && lck1->fnum == lck2->fnum) { + return False; + } + + if (lck2->start == 0 && lck2->size == 0 && lck1->size != 0) { + return True; + } + + if (lck1->start >= (lck2->start + lck2->size) || + lck2->start >= (lck1->start + lck1->size)) { + return False; + } + + return True; +} +#endif + +/**************************************************************************** + Check to see if this lock conflicts, but ignore our own locks on the + same fnum only. This is the read/write lock check code path. + This is never used in the POSIX lock case. +****************************************************************************/ + +static bool brl_conflict_other(const struct lock_struct *lock, + const struct lock_struct *rw_probe) +{ + if (lock->lock_type == READ_LOCK && rw_probe->lock_type == READ_LOCK) { + return False; + } + + if (lock->lock_flav == POSIX_LOCK && + rw_probe->lock_flav == POSIX_LOCK) { + /* + * POSIX flavour locks never conflict here - this is only called + * in the read/write path. + */ + return False; + } + + if (!brl_overlap(lock, rw_probe)) { + /* + * I/O can only conflict when overlapping a lock, thus let it + * pass + */ + return false; + } + + if (!brl_same_context(&lock->context, &rw_probe->context)) { + /* + * Different process, conflict + */ + return true; + } + + if (lock->fnum != rw_probe->fnum) { + /* + * Different file handle, conflict + */ + return true; + } + + if ((lock->lock_type == READ_LOCK) && + (rw_probe->lock_type == WRITE_LOCK)) { + /* + * Incoming WRITE locks conflict with existing READ locks even + * if the context is the same. JRA. See LOCKTEST7 in + * smbtorture. + */ + return true; + } + + /* + * I/O request compatible with existing lock, let it pass without + * conflict + */ + + return false; +} + +/**************************************************************************** + Open up the brlock.tdb database. +****************************************************************************/ + +void brl_init(bool read_only) +{ + int tdb_flags; + char *db_path; + + if (brlock_db) { + return; + } + + tdb_flags = + TDB_DEFAULT| + TDB_VOLATILE| + TDB_CLEAR_IF_FIRST| + TDB_INCOMPATIBLE_HASH| + TDB_SEQNUM; + + db_path = lock_path(talloc_tos(), "brlock.tdb"); + if (db_path == NULL) { + DEBUG(0, ("out of memory!\n")); + return; + } + + brlock_db = db_open(NULL, db_path, + SMB_OPEN_DATABASE_TDB_HASH_SIZE, tdb_flags, + read_only?O_RDONLY:(O_RDWR|O_CREAT), 0644, + DBWRAP_LOCK_ORDER_2, DBWRAP_FLAG_NONE); + if (!brlock_db) { + DEBUG(0,("Failed to open byte range locking database %s\n", + db_path)); + TALLOC_FREE(db_path); + return; + } + TALLOC_FREE(db_path); +} + +/**************************************************************************** + Close down the brlock.tdb database. +****************************************************************************/ + +void brl_shutdown(void) +{ + TALLOC_FREE(brlock_db); +} + +#if ZERO_ZERO +/**************************************************************************** + Compare two locks for sorting. +****************************************************************************/ + +static int lock_compare(const struct lock_struct *lck1, + const struct lock_struct *lck2) +{ + if (lck1->start != lck2->start) { + return (lck1->start - lck2->start); + } + if (lck2->size != lck1->size) { + return ((int)lck1->size - (int)lck2->size); + } + return 0; +} +#endif + +/**************************************************************************** + Lock a range of bytes - Windows lock semantics. +****************************************************************************/ + +NTSTATUS brl_lock_windows_default(struct byte_range_lock *br_lck, + struct lock_struct *plock) +{ + unsigned int i; + files_struct *fsp = br_lck->fsp; + struct lock_struct *locks = br_lck->lock_data; + NTSTATUS status; + bool valid; + + SMB_ASSERT(plock->lock_type != UNLOCK_LOCK); + + valid = byte_range_valid(plock->start, plock->size); + if (!valid) { + return NT_STATUS_INVALID_LOCK_RANGE; + } + + for (i=0; i < br_lck->num_locks; i++) { + /* Do any Windows or POSIX locks conflict ? */ + if (brl_conflict(&locks[i], plock)) { + if (!serverid_exists(&locks[i].context.pid)) { + locks[i].context.pid.pid = 0; + br_lck->modified = true; + continue; + } + /* Remember who blocked us. */ + plock->context.smblctx = locks[i].context.smblctx; + return NT_STATUS_LOCK_NOT_GRANTED; + } +#if ZERO_ZERO + if (plock->start == 0 && plock->size == 0 && + locks[i].size == 0) { + break; + } +#endif + } + + contend_level2_oplocks_begin(fsp, LEVEL2_CONTEND_WINDOWS_BRL); + + /* We can get the Windows lock, now see if it needs to + be mapped into a lower level POSIX one, and if so can + we get it ? */ + + if (lp_posix_locking(fsp->conn->params)) { + int errno_ret; + if (!set_posix_lock_windows_flavour(fsp, + plock->start, + plock->size, + plock->lock_type, + &plock->context, + locks, + br_lck->num_locks, + &errno_ret)) { + + /* We don't know who blocked us. */ + plock->context.smblctx = 0xFFFFFFFFFFFFFFFFLL; + + if (errno_ret == EACCES || errno_ret == EAGAIN) { + status = NT_STATUS_LOCK_NOT_GRANTED; + goto fail; + } else { + status = map_nt_error_from_unix(errno); + goto fail; + } + } + } + + /* no conflicts - add it to the list of locks */ + locks = talloc_realloc(br_lck, locks, struct lock_struct, + (br_lck->num_locks + 1)); + if (!locks) { + status = NT_STATUS_NO_MEMORY; + goto fail; + } + + memcpy(&locks[br_lck->num_locks], plock, sizeof(struct lock_struct)); + br_lck->num_locks += 1; + br_lck->lock_data = locks; + br_lck->modified = True; + + return NT_STATUS_OK; + fail: + contend_level2_oplocks_end(fsp, LEVEL2_CONTEND_WINDOWS_BRL); + return status; +} + +/**************************************************************************** + Cope with POSIX range splits and merges. +****************************************************************************/ + +static unsigned int brlock_posix_split_merge(struct lock_struct *lck_arr, /* Output array. */ + struct lock_struct *ex, /* existing lock. */ + struct lock_struct *plock) /* proposed lock. */ +{ + bool lock_types_differ = (ex->lock_type != plock->lock_type); + + /* We can't merge non-conflicting locks on different context - ignore fnum. */ + + if (!brl_same_context(&ex->context, &plock->context)) { + /* Just copy. */ + memcpy(&lck_arr[0], ex, sizeof(struct lock_struct)); + return 1; + } + + /* We now know we have the same context. */ + + /* Did we overlap ? */ + +/********************************************* + +---------+ + | ex | + +---------+ + +-------+ + | plock | + +-------+ +OR.... + +---------+ + | ex | + +---------+ +**********************************************/ + + if ( (ex->start > (plock->start + plock->size)) || + (plock->start > (ex->start + ex->size))) { + + /* No overlap with this lock - copy existing. */ + + memcpy(&lck_arr[0], ex, sizeof(struct lock_struct)); + return 1; + } + +/********************************************* + +---------------------------+ + | ex | + +---------------------------+ + +---------------------------+ + | plock | -> replace with plock. + +---------------------------+ +OR + +---------------+ + | ex | + +---------------+ + +---------------------------+ + | plock | -> replace with plock. + +---------------------------+ + +**********************************************/ + + if ( (ex->start >= plock->start) && + (ex->start + ex->size <= plock->start + plock->size) ) { + + /* Replace - discard existing lock. */ + + return 0; + } + +/********************************************* +Adjacent after. + +-------+ + | ex | + +-------+ + +---------------+ + | plock | + +---------------+ + +BECOMES.... + +---------------+-------+ + | plock | ex | - different lock types. + +---------------+-------+ +OR.... (merge) + +-----------------------+ + | plock | - same lock type. + +-----------------------+ +**********************************************/ + + if (plock->start + plock->size == ex->start) { + + /* If the lock types are the same, we merge, if different, we + add the remainder of the old lock. */ + + if (lock_types_differ) { + /* Add existing. */ + memcpy(&lck_arr[0], ex, sizeof(struct lock_struct)); + return 1; + } else { + /* Merge - adjust incoming lock as we may have more + * merging to come. */ + plock->size += ex->size; + return 0; + } + } + +/********************************************* +Adjacent before. + +-------+ + | ex | + +-------+ + +---------------+ + | plock | + +---------------+ +BECOMES.... + +-------+---------------+ + | ex | plock | - different lock types + +-------+---------------+ + +OR.... (merge) + +-----------------------+ + | plock | - same lock type. + +-----------------------+ + +**********************************************/ + + if (ex->start + ex->size == plock->start) { + + /* If the lock types are the same, we merge, if different, we + add the existing lock. */ + + if (lock_types_differ) { + memcpy(&lck_arr[0], ex, sizeof(struct lock_struct)); + return 1; + } else { + /* Merge - adjust incoming lock as we may have more + * merging to come. */ + plock->start = ex->start; + plock->size += ex->size; + return 0; + } + } + +/********************************************* +Overlap after. + +-----------------------+ + | ex | + +-----------------------+ + +---------------+ + | plock | + +---------------+ +OR + +----------------+ + | ex | + +----------------+ + +---------------+ + | plock | + +---------------+ + +BECOMES.... + +---------------+-------+ + | plock | ex | - different lock types. + +---------------+-------+ +OR.... (merge) + +-----------------------+ + | plock | - same lock type. + +-----------------------+ +**********************************************/ + + if ( (ex->start >= plock->start) && + (ex->start <= plock->start + plock->size) && + (ex->start + ex->size > plock->start + plock->size) ) { + + /* If the lock types are the same, we merge, if different, we + add the remainder of the old lock. */ + + if (lock_types_differ) { + /* Add remaining existing. */ + memcpy(&lck_arr[0], ex, sizeof(struct lock_struct)); + /* Adjust existing start and size. */ + lck_arr[0].start = plock->start + plock->size; + lck_arr[0].size = (ex->start + ex->size) - (plock->start + plock->size); + return 1; + } else { + /* Merge - adjust incoming lock as we may have more + * merging to come. */ + plock->size += (ex->start + ex->size) - (plock->start + plock->size); + return 0; + } + } + +/********************************************* +Overlap before. + +-----------------------+ + | ex | + +-----------------------+ + +---------------+ + | plock | + +---------------+ +OR + +-------------+ + | ex | + +-------------+ + +---------------+ + | plock | + +---------------+ + +BECOMES.... + +-------+---------------+ + | ex | plock | - different lock types + +-------+---------------+ + +OR.... (merge) + +-----------------------+ + | plock | - same lock type. + +-----------------------+ + +**********************************************/ + + if ( (ex->start < plock->start) && + (ex->start + ex->size >= plock->start) && + (ex->start + ex->size <= plock->start + plock->size) ) { + + /* If the lock types are the same, we merge, if different, we + add the truncated old lock. */ + + if (lock_types_differ) { + memcpy(&lck_arr[0], ex, sizeof(struct lock_struct)); + /* Adjust existing size. */ + lck_arr[0].size = plock->start - ex->start; + return 1; + } else { + /* Merge - adjust incoming lock as we may have more + * merging to come. MUST ADJUST plock SIZE FIRST ! */ + plock->size += (plock->start - ex->start); + plock->start = ex->start; + return 0; + } + } + +/********************************************* +Complete overlap. + +---------------------------+ + | ex | + +---------------------------+ + +---------+ + | plock | + +---------+ +BECOMES..... + +-------+---------+---------+ + | ex | plock | ex | - different lock types. + +-------+---------+---------+ +OR + +---------------------------+ + | plock | - same lock type. + +---------------------------+ +**********************************************/ + + if ( (ex->start < plock->start) && (ex->start + ex->size > plock->start + plock->size) ) { + + if (lock_types_differ) { + + /* We have to split ex into two locks here. */ + + memcpy(&lck_arr[0], ex, sizeof(struct lock_struct)); + memcpy(&lck_arr[1], ex, sizeof(struct lock_struct)); + + /* Adjust first existing size. */ + lck_arr[0].size = plock->start - ex->start; + + /* Adjust second existing start and size. */ + lck_arr[1].start = plock->start + plock->size; + lck_arr[1].size = (ex->start + ex->size) - (plock->start + plock->size); + return 2; + } else { + /* Just eat the existing locks, merge them into plock. */ + plock->start = ex->start; + plock->size = ex->size; + return 0; + } + } + + /* Never get here. */ + smb_panic("brlock_posix_split_merge"); + /* Notreached. */ + + /* Keep some compilers happy. */ + return 0; +} + +/**************************************************************************** + Lock a range of bytes - POSIX lock semantics. + We must cope with range splits and merges. +****************************************************************************/ + +static NTSTATUS brl_lock_posix(struct byte_range_lock *br_lck, + struct lock_struct *plock) +{ + unsigned int i, count, posix_count; + struct lock_struct *locks = br_lck->lock_data; + struct lock_struct *tp; + bool break_oplocks = false; + NTSTATUS status; + + /* No zero-zero locks for POSIX. */ + if (plock->start == 0 && plock->size == 0) { + return NT_STATUS_INVALID_PARAMETER; + } + + /* Don't allow 64-bit lock wrap. */ + if (plock->start + plock->size - 1 < plock->start) { + return NT_STATUS_INVALID_PARAMETER; + } + + /* The worst case scenario here is we have to split an + existing POSIX lock range into two, and add our lock, + so we need at most 2 more entries. */ + + tp = talloc_array(br_lck, struct lock_struct, br_lck->num_locks + 2); + if (!tp) { + return NT_STATUS_NO_MEMORY; + } + + count = posix_count = 0; + + for (i=0; i < br_lck->num_locks; i++) { + struct lock_struct *curr_lock = &locks[i]; + + if (curr_lock->lock_flav == WINDOWS_LOCK) { + /* Do any Windows flavour locks conflict ? */ + if (brl_conflict(curr_lock, plock)) { + if (!serverid_exists(&curr_lock->context.pid)) { + curr_lock->context.pid.pid = 0; + br_lck->modified = true; + continue; + } + /* No games with error messages. */ + TALLOC_FREE(tp); + /* Remember who blocked us. */ + plock->context.smblctx = curr_lock->context.smblctx; + return NT_STATUS_LOCK_NOT_GRANTED; + } + /* Just copy the Windows lock into the new array. */ + memcpy(&tp[count], curr_lock, sizeof(struct lock_struct)); + count++; + } else { + unsigned int tmp_count = 0; + + /* POSIX conflict semantics are different. */ + if (brl_conflict_posix(curr_lock, plock)) { + if (!serverid_exists(&curr_lock->context.pid)) { + curr_lock->context.pid.pid = 0; + br_lck->modified = true; + continue; + } + /* Can't block ourselves with POSIX locks. */ + /* No games with error messages. */ + TALLOC_FREE(tp); + /* Remember who blocked us. */ + plock->context.smblctx = curr_lock->context.smblctx; + return NT_STATUS_LOCK_NOT_GRANTED; + } + + /* Work out overlaps. */ + tmp_count += brlock_posix_split_merge(&tp[count], curr_lock, plock); + posix_count += tmp_count; + count += tmp_count; + } + } + + /* + * Break oplocks while we hold a brl. Since lock() and unlock() calls + * are not symetric with POSIX semantics, we cannot guarantee our + * contend_level2_oplocks_begin/end calls will be acquired and + * released one-for-one as with Windows semantics. Therefore we only + * call contend_level2_oplocks_begin if this is the first POSIX brl on + * the file. + */ + break_oplocks = (posix_count == 0); + if (break_oplocks) { + contend_level2_oplocks_begin(br_lck->fsp, + LEVEL2_CONTEND_POSIX_BRL); + } + + /* Try and add the lock in order, sorted by lock start. */ + for (i=0; i < count; i++) { + struct lock_struct *curr_lock = &tp[i]; + + if (curr_lock->start <= plock->start) { + continue; + } + } + + if (i < count) { + memmove(&tp[i+1], &tp[i], + (count - i)*sizeof(struct lock_struct)); + } + memcpy(&tp[i], plock, sizeof(struct lock_struct)); + count++; + + /* We can get the POSIX lock, now see if it needs to + be mapped into a lower level POSIX one, and if so can + we get it ? */ + + if (lp_posix_locking(br_lck->fsp->conn->params)) { + int errno_ret; + + /* The lower layer just needs to attempt to + get the system POSIX lock. We've weeded out + any conflicts above. */ + + if (!set_posix_lock_posix_flavour(br_lck->fsp, + plock->start, + plock->size, + plock->lock_type, + &plock->context, + &errno_ret)) { + + /* We don't know who blocked us. */ + plock->context.smblctx = 0xFFFFFFFFFFFFFFFFLL; + + if (errno_ret == EACCES || errno_ret == EAGAIN) { + TALLOC_FREE(tp); + status = NT_STATUS_LOCK_NOT_GRANTED; + goto fail; + } else { + TALLOC_FREE(tp); + status = map_nt_error_from_unix(errno); + goto fail; + } + } + } + + /* If we didn't use all the allocated size, + * Realloc so we don't leak entries per lock call. */ + if (count < br_lck->num_locks + 2) { + tp = talloc_realloc(br_lck, tp, struct lock_struct, count); + if (!tp) { + status = NT_STATUS_NO_MEMORY; + goto fail; + } + } + + br_lck->num_locks = count; + TALLOC_FREE(br_lck->lock_data); + br_lck->lock_data = tp; + locks = tp; + br_lck->modified = True; + + /* A successful downgrade from write to read lock can trigger a lock + re-evalutation where waiting readers can now proceed. */ + + return NT_STATUS_OK; + fail: + if (break_oplocks) { + contend_level2_oplocks_end(br_lck->fsp, + LEVEL2_CONTEND_POSIX_BRL); + } + return status; +} + +NTSTATUS smb_vfs_call_brl_lock_windows(struct vfs_handle_struct *handle, + struct byte_range_lock *br_lck, + struct lock_struct *plock) +{ + VFS_FIND(brl_lock_windows); + return handle->fns->brl_lock_windows_fn(handle, br_lck, plock); +} + +/**************************************************************************** + Lock a range of bytes. +****************************************************************************/ + +NTSTATUS brl_lock( + struct byte_range_lock *br_lck, + uint64_t smblctx, + struct server_id pid, + br_off start, + br_off size, + enum brl_type lock_type, + enum brl_flavour lock_flav, + struct server_id *blocker_pid, + uint64_t *psmblctx) +{ + NTSTATUS ret; + struct lock_struct lock; + + ZERO_STRUCT(lock); + +#if !ZERO_ZERO + if (start == 0 && size == 0) { + DEBUG(0,("client sent 0/0 lock - please report this\n")); + } +#endif + + lock = (struct lock_struct) { + .context.smblctx = smblctx, + .context.pid = pid, + .context.tid = br_lck->fsp->conn->cnum, + .start = start, + .size = size, + .fnum = br_lck->fsp->fnum, + .lock_type = lock_type, + .lock_flav = lock_flav + }; + + if (lock_flav == WINDOWS_LOCK) { + ret = SMB_VFS_BRL_LOCK_WINDOWS( + br_lck->fsp->conn, br_lck, &lock); + } else { + ret = brl_lock_posix(br_lck, &lock); + } + +#if ZERO_ZERO + /* sort the lock list */ + TYPESAFE_QSORT(br_lck->lock_data, (size_t)br_lck->num_locks, lock_compare); +#endif + /* If we're returning an error, return who blocked us. */ + if (!NT_STATUS_IS_OK(ret) && psmblctx) { + *blocker_pid = lock.context.pid; + *psmblctx = lock.context.smblctx; + } + return ret; +} + +/**************************************************************************** + Unlock a range of bytes - Windows semantics. +****************************************************************************/ + +bool brl_unlock_windows_default(struct byte_range_lock *br_lck, + const struct lock_struct *plock) +{ + unsigned int i; + struct lock_struct *locks = br_lck->lock_data; + enum brl_type deleted_lock_type = READ_LOCK; /* shut the compiler up.... */ + + SMB_ASSERT(plock->lock_type == UNLOCK_LOCK); + +#if ZERO_ZERO + /* Delete write locks by preference... The lock list + is sorted in the zero zero case. */ + + for (i = 0; i < br_lck->num_locks; i++) { + struct lock_struct *lock = &locks[i]; + + if (lock->lock_type == WRITE_LOCK && + brl_same_context(&lock->context, &plock->context) && + lock->fnum == plock->fnum && + lock->lock_flav == WINDOWS_LOCK && + lock->start == plock->start && + lock->size == plock->size) { + + /* found it - delete it */ + deleted_lock_type = lock->lock_type; + break; + } + } + + if (i != br_lck->num_locks) { + /* We found it - don't search again. */ + goto unlock_continue; + } +#endif + + for (i = 0; i < br_lck->num_locks; i++) { + struct lock_struct *lock = &locks[i]; + + /* Only remove our own locks that match in start, size, and flavour. */ + if (brl_same_context(&lock->context, &plock->context) && + lock->fnum == plock->fnum && + lock->lock_flav == WINDOWS_LOCK && + lock->start == plock->start && + lock->size == plock->size ) { + deleted_lock_type = lock->lock_type; + break; + } + } + + if (i == br_lck->num_locks) { + /* we didn't find it */ + return False; + } + +#if ZERO_ZERO + unlock_continue: +#endif + + ARRAY_DEL_ELEMENT(locks, i, br_lck->num_locks); + br_lck->num_locks -= 1; + br_lck->modified = True; + + /* Unlock the underlying POSIX regions. */ + if(lp_posix_locking(br_lck->fsp->conn->params)) { + release_posix_lock_windows_flavour(br_lck->fsp, + plock->start, + plock->size, + deleted_lock_type, + &plock->context, + locks, + br_lck->num_locks); + } + + contend_level2_oplocks_end(br_lck->fsp, LEVEL2_CONTEND_WINDOWS_BRL); + return True; +} + +/**************************************************************************** + Unlock a range of bytes - POSIX semantics. +****************************************************************************/ + +static bool brl_unlock_posix(struct byte_range_lock *br_lck, + struct lock_struct *plock) +{ + unsigned int i, count; + struct lock_struct *tp; + struct lock_struct *locks = br_lck->lock_data; + bool overlap_found = False; + + /* No zero-zero locks for POSIX. */ + if (plock->start == 0 && plock->size == 0) { + return False; + } + + /* Don't allow 64-bit lock wrap. */ + if (plock->start + plock->size < plock->start || + plock->start + plock->size < plock->size) { + DEBUG(10,("brl_unlock_posix: lock wrap\n")); + return False; + } + + /* The worst case scenario here is we have to split an + existing POSIX lock range into two, so we need at most + 1 more entry. */ + + tp = talloc_array(br_lck, struct lock_struct, br_lck->num_locks + 1); + if (!tp) { + DEBUG(10,("brl_unlock_posix: malloc fail\n")); + return False; + } + + count = 0; + for (i = 0; i < br_lck->num_locks; i++) { + struct lock_struct *lock = &locks[i]; + unsigned int tmp_count; + + /* Only remove our own locks - ignore fnum. */ + if (!brl_same_context(&lock->context, &plock->context)) { + memcpy(&tp[count], lock, sizeof(struct lock_struct)); + count++; + continue; + } + + if (lock->lock_flav == WINDOWS_LOCK) { + /* Do any Windows flavour locks conflict ? */ + if (brl_conflict(lock, plock)) { + TALLOC_FREE(tp); + return false; + } + /* Just copy the Windows lock into the new array. */ + memcpy(&tp[count], lock, sizeof(struct lock_struct)); + count++; + continue; + } + + /* Work out overlaps. */ + tmp_count = brlock_posix_split_merge(&tp[count], lock, plock); + + if (tmp_count == 0) { + /* plock overlapped the existing lock completely, + or replaced it. Don't copy the existing lock. */ + overlap_found = true; + } else if (tmp_count == 1) { + /* Either no overlap, (simple copy of existing lock) or + * an overlap of an existing lock. */ + /* If the lock changed size, we had an overlap. */ + if (tp[count].size != lock->size) { + overlap_found = true; + } + count += tmp_count; + } else if (tmp_count == 2) { + /* We split a lock range in two. */ + overlap_found = true; + count += tmp_count; + + /* Optimisation... */ + /* We know we're finished here as we can't overlap any + more POSIX locks. Copy the rest of the lock array. */ + + if (i < br_lck->num_locks - 1) { + memcpy(&tp[count], &locks[i+1], + sizeof(*locks)*((br_lck->num_locks-1) - i)); + count += ((br_lck->num_locks-1) - i); + } + break; + } + + } + + if (!overlap_found) { + /* Just ignore - no change. */ + TALLOC_FREE(tp); + DEBUG(10,("brl_unlock_posix: No overlap - unlocked.\n")); + return True; + } + + /* Unlock any POSIX regions. */ + if(lp_posix_locking(br_lck->fsp->conn->params)) { + release_posix_lock_posix_flavour(br_lck->fsp, + plock->start, + plock->size, + &plock->context, + tp, + count); + } + + /* Realloc so we don't leak entries per unlock call. */ + if (count) { + tp = talloc_realloc(br_lck, tp, struct lock_struct, count); + if (!tp) { + DEBUG(10,("brl_unlock_posix: realloc fail\n")); + return False; + } + } else { + /* We deleted the last lock. */ + TALLOC_FREE(tp); + tp = NULL; + } + + contend_level2_oplocks_end(br_lck->fsp, + LEVEL2_CONTEND_POSIX_BRL); + + br_lck->num_locks = count; + TALLOC_FREE(br_lck->lock_data); + locks = tp; + br_lck->lock_data = tp; + br_lck->modified = True; + + return True; +} + +bool smb_vfs_call_brl_unlock_windows(struct vfs_handle_struct *handle, + struct byte_range_lock *br_lck, + const struct lock_struct *plock) +{ + VFS_FIND(brl_unlock_windows); + return handle->fns->brl_unlock_windows_fn(handle, br_lck, plock); +} + +/**************************************************************************** + Unlock a range of bytes. +****************************************************************************/ + +bool brl_unlock(struct byte_range_lock *br_lck, + uint64_t smblctx, + struct server_id pid, + br_off start, + br_off size, + enum brl_flavour lock_flav) +{ + struct lock_struct lock; + + lock.context.smblctx = smblctx; + lock.context.pid = pid; + lock.context.tid = br_lck->fsp->conn->cnum; + lock.start = start; + lock.size = size; + lock.fnum = br_lck->fsp->fnum; + lock.lock_type = UNLOCK_LOCK; + lock.lock_flav = lock_flav; + + if (lock_flav == WINDOWS_LOCK) { + return SMB_VFS_BRL_UNLOCK_WINDOWS( + br_lck->fsp->conn, br_lck, &lock); + } else { + return brl_unlock_posix(br_lck, &lock); + } +} + +/**************************************************************************** + Test if we could add a lock if we wanted to. + Returns True if the region required is currently unlocked, False if locked. +****************************************************************************/ + +bool brl_locktest(struct byte_range_lock *br_lck, + const struct lock_struct *rw_probe) +{ + bool ret = True; + unsigned int i; + struct lock_struct *locks = br_lck->lock_data; + files_struct *fsp = br_lck->fsp; + + /* Make sure existing locks don't conflict */ + for (i=0; i < br_lck->num_locks; i++) { + /* + * Our own locks don't conflict. + */ + if (brl_conflict_other(&locks[i], rw_probe)) { + if (br_lck->record == NULL) { + /* readonly */ + return false; + } + + if (!serverid_exists(&locks[i].context.pid)) { + locks[i].context.pid.pid = 0; + br_lck->modified = true; + continue; + } + + return False; + } + } + + /* + * There is no lock held by an SMB daemon, check to + * see if there is a POSIX lock from a UNIX or NFS process. + * This only conflicts with Windows locks, not POSIX locks. + */ + + if(lp_posix_locking(fsp->conn->params) && + (rw_probe->lock_flav == WINDOWS_LOCK)) { + /* + * Make copies -- is_posix_locked might modify the values + */ + + br_off start = rw_probe->start; + br_off size = rw_probe->size; + enum brl_type lock_type = rw_probe->lock_type; + + ret = is_posix_locked(fsp, &start, &size, &lock_type, WINDOWS_LOCK); + + DEBUG(10, ("brl_locktest: posix start=%ju len=%ju %s for %s " + "file %s\n", (uintmax_t)start, (uintmax_t)size, + ret ? "locked" : "unlocked", + fsp_fnum_dbg(fsp), fsp_str_dbg(fsp))); + + /* We need to return the inverse of is_posix_locked. */ + ret = !ret; + } + + /* no conflicts - we could have added it */ + return ret; +} + +/**************************************************************************** + Query for existing locks. +****************************************************************************/ + +NTSTATUS brl_lockquery(struct byte_range_lock *br_lck, + uint64_t *psmblctx, + struct server_id pid, + br_off *pstart, + br_off *psize, + enum brl_type *plock_type, + enum brl_flavour lock_flav) +{ + unsigned int i; + struct lock_struct lock; + const struct lock_struct *locks = br_lck->lock_data; + files_struct *fsp = br_lck->fsp; + + lock.context.smblctx = *psmblctx; + lock.context.pid = pid; + lock.context.tid = br_lck->fsp->conn->cnum; + lock.start = *pstart; + lock.size = *psize; + lock.fnum = fsp->fnum; + lock.lock_type = *plock_type; + lock.lock_flav = lock_flav; + + /* Make sure existing locks don't conflict */ + for (i=0; i < br_lck->num_locks; i++) { + const struct lock_struct *exlock = &locks[i]; + bool conflict = False; + + if (exlock->lock_flav == WINDOWS_LOCK) { + conflict = brl_conflict(exlock, &lock); + } else { + conflict = brl_conflict_posix(exlock, &lock); + } + + if (conflict) { + *psmblctx = exlock->context.smblctx; + *pstart = exlock->start; + *psize = exlock->size; + *plock_type = exlock->lock_type; + return NT_STATUS_LOCK_NOT_GRANTED; + } + } + + /* + * There is no lock held by an SMB daemon, check to + * see if there is a POSIX lock from a UNIX or NFS process. + */ + + if(lp_posix_locking(fsp->conn->params)) { + bool ret = is_posix_locked(fsp, pstart, psize, plock_type, POSIX_LOCK); + + DEBUG(10, ("brl_lockquery: posix start=%ju len=%ju %s for %s " + "file %s\n", (uintmax_t)*pstart, + (uintmax_t)*psize, ret ? "locked" : "unlocked", + fsp_fnum_dbg(fsp), fsp_str_dbg(fsp))); + + if (ret) { + /* Hmmm. No clue what to set smblctx to - use -1. */ + *psmblctx = 0xFFFFFFFFFFFFFFFFLL; + return NT_STATUS_LOCK_NOT_GRANTED; + } + } + + return NT_STATUS_OK; +} + + +/**************************************************************************** + Remove any locks associated with a open file. + We return True if this process owns any other Windows locks on this + fd and so we should not immediately close the fd. +****************************************************************************/ + +void brl_close_fnum(struct byte_range_lock *br_lck) +{ + files_struct *fsp = br_lck->fsp; + uint32_t tid = fsp->conn->cnum; + uint64_t fnum = fsp->fnum; + unsigned int i; + struct lock_struct *locks = br_lck->lock_data; + struct server_id pid = messaging_server_id(fsp->conn->sconn->msg_ctx); + struct lock_struct *locks_copy; + unsigned int num_locks_copy; + + /* Copy the current lock array. */ + if (br_lck->num_locks) { + locks_copy = (struct lock_struct *)talloc_memdup(br_lck, locks, br_lck->num_locks * sizeof(struct lock_struct)); + if (!locks_copy) { + smb_panic("brl_close_fnum: talloc failed"); + } + } else { + locks_copy = NULL; + } + + num_locks_copy = br_lck->num_locks; + + for (i=0; i < num_locks_copy; i++) { + struct lock_struct *lock = &locks_copy[i]; + + if (lock->context.tid == tid && + server_id_equal(&lock->context.pid, &pid) && + (lock->fnum == fnum)) { + brl_unlock( + br_lck, + lock->context.smblctx, + pid, + lock->start, + lock->size, + lock->lock_flav); + } + } +} + +bool brl_mark_disconnected(struct files_struct *fsp) +{ + uint32_t tid = fsp->conn->cnum; + uint64_t smblctx; + uint64_t fnum = fsp->fnum; + unsigned int i; + struct server_id self = messaging_server_id(fsp->conn->sconn->msg_ctx); + struct byte_range_lock *br_lck = NULL; + + if (fsp->op == NULL) { + return false; + } + + smblctx = fsp->op->global->open_persistent_id; + + if (!fsp->op->global->durable) { + return false; + } + + if (fsp->current_lock_count == 0) { + return true; + } + + br_lck = brl_get_locks(talloc_tos(), fsp); + if (br_lck == NULL) { + return false; + } + + for (i=0; i < br_lck->num_locks; i++) { + struct lock_struct *lock = &br_lck->lock_data[i]; + + /* + * as this is a durable handle, we only expect locks + * of the current file handle! + */ + + if (lock->context.smblctx != smblctx) { + TALLOC_FREE(br_lck); + return false; + } + + if (lock->context.tid != tid) { + TALLOC_FREE(br_lck); + return false; + } + + if (!server_id_equal(&lock->context.pid, &self)) { + TALLOC_FREE(br_lck); + return false; + } + + if (lock->fnum != fnum) { + TALLOC_FREE(br_lck); + return false; + } + + server_id_set_disconnected(&lock->context.pid); + lock->context.tid = TID_FIELD_INVALID; + lock->fnum = FNUM_FIELD_INVALID; + } + + br_lck->modified = true; + TALLOC_FREE(br_lck); + return true; +} + +bool brl_reconnect_disconnected(struct files_struct *fsp) +{ + uint32_t tid = fsp->conn->cnum; + uint64_t smblctx; + uint64_t fnum = fsp->fnum; + unsigned int i; + struct server_id self = messaging_server_id(fsp->conn->sconn->msg_ctx); + struct byte_range_lock *br_lck = NULL; + + if (fsp->op == NULL) { + return false; + } + + smblctx = fsp->op->global->open_persistent_id; + + if (!fsp->op->global->durable) { + return false; + } + + /* + * When reconnecting, we do not want to validate the brlock entries + * and thereby remove our own (disconnected) entries but reactivate + * them instead. + */ + + br_lck = brl_get_locks(talloc_tos(), fsp); + if (br_lck == NULL) { + return false; + } + + if (br_lck->num_locks == 0) { + TALLOC_FREE(br_lck); + return true; + } + + for (i=0; i < br_lck->num_locks; i++) { + struct lock_struct *lock = &br_lck->lock_data[i]; + + /* + * as this is a durable handle we only expect locks + * of the current file handle! + */ + + if (lock->context.smblctx != smblctx) { + TALLOC_FREE(br_lck); + return false; + } + + if (lock->context.tid != TID_FIELD_INVALID) { + TALLOC_FREE(br_lck); + return false; + } + + if (!server_id_is_disconnected(&lock->context.pid)) { + TALLOC_FREE(br_lck); + return false; + } + + if (lock->fnum != FNUM_FIELD_INVALID) { + TALLOC_FREE(br_lck); + return false; + } + + lock->context.pid = self; + lock->context.tid = tid; + lock->fnum = fnum; + } + + fsp->current_lock_count = br_lck->num_locks; + br_lck->modified = true; + TALLOC_FREE(br_lck); + return true; +} + +struct brl_forall_cb { + void (*fn)(struct file_id id, struct server_id pid, + enum brl_type lock_type, + enum brl_flavour lock_flav, + br_off start, br_off size, + void *private_data); + void *private_data; +}; + +/**************************************************************************** + Traverse the whole database with this function, calling traverse_callback + on each lock. +****************************************************************************/ + +static int brl_traverse_fn(struct db_record *rec, void *state) +{ + struct brl_forall_cb *cb = (struct brl_forall_cb *)state; + struct lock_struct *locks; + struct file_id *key; + unsigned int i; + unsigned int num_locks = 0; + TDB_DATA dbkey; + TDB_DATA value; + + dbkey = dbwrap_record_get_key(rec); + value = dbwrap_record_get_value(rec); + + /* In a traverse function we must make a copy of + dbuf before modifying it. */ + + locks = (struct lock_struct *)talloc_memdup( + talloc_tos(), value.dptr, value.dsize); + if (!locks) { + return -1; /* Terminate traversal. */ + } + + key = (struct file_id *)dbkey.dptr; + num_locks = value.dsize/sizeof(*locks); + + if (cb->fn) { + for ( i=0; i<num_locks; i++) { + cb->fn(*key, + locks[i].context.pid, + locks[i].lock_type, + locks[i].lock_flav, + locks[i].start, + locks[i].size, + cb->private_data); + } + } + + TALLOC_FREE(locks); + return 0; +} + +/******************************************************************* + Call the specified function on each lock in the database. +********************************************************************/ + +int brl_forall(void (*fn)(struct file_id id, struct server_id pid, + enum brl_type lock_type, + enum brl_flavour lock_flav, + br_off start, br_off size, + void *private_data), + void *private_data) +{ + struct brl_forall_cb cb; + NTSTATUS status; + int count = 0; + + if (!brlock_db) { + return 0; + } + cb.fn = fn; + cb.private_data = private_data; + status = dbwrap_traverse(brlock_db, brl_traverse_fn, &cb, &count); + + if (!NT_STATUS_IS_OK(status)) { + return -1; + } else { + return count; + } +} + +/******************************************************************* + Store a potentially modified set of byte range lock data back into + the database. + Unlock the record. +********************************************************************/ + +static void byte_range_lock_flush(struct byte_range_lock *br_lck) +{ + unsigned i; + struct lock_struct *locks = br_lck->lock_data; + + if (!br_lck->modified) { + DEBUG(10, ("br_lck not modified\n")); + goto done; + } + + i = 0; + + while (i < br_lck->num_locks) { + if (locks[i].context.pid.pid == 0) { + /* + * Autocleanup, the process conflicted and does not + * exist anymore. + */ + locks[i] = locks[br_lck->num_locks-1]; + br_lck->num_locks -= 1; + } else { + i += 1; + } + } + + if (br_lck->num_locks == 0) { + /* No locks - delete this entry. */ + NTSTATUS status = dbwrap_record_delete(br_lck->record); + if (!NT_STATUS_IS_OK(status)) { + DEBUG(0, ("delete_rec returned %s\n", + nt_errstr(status))); + smb_panic("Could not delete byte range lock entry"); + } + } else { + TDB_DATA data = { + .dsize = br_lck->num_locks * sizeof(struct lock_struct), + .dptr = (uint8_t *)br_lck->lock_data, + }; + NTSTATUS status; + + status = dbwrap_record_store(br_lck->record, data, TDB_REPLACE); + if (!NT_STATUS_IS_OK(status)) { + DEBUG(0, ("store returned %s\n", nt_errstr(status))); + smb_panic("Could not store byte range mode entry"); + } + } + + DEBUG(10, ("seqnum=%d\n", dbwrap_get_seqnum(brlock_db))); + + done: + br_lck->modified = false; + TALLOC_FREE(br_lck->record); +} + +static int byte_range_lock_destructor(struct byte_range_lock *br_lck) +{ + byte_range_lock_flush(br_lck); + return 0; +} + +static bool brl_parse_data(struct byte_range_lock *br_lck, TDB_DATA data) +{ + size_t data_len; + + if (data.dsize == 0) { + return true; + } + if (data.dsize % sizeof(struct lock_struct) != 0) { + DEBUG(1, ("Invalid data size: %u\n", (unsigned)data.dsize)); + return false; + } + + br_lck->num_locks = data.dsize / sizeof(struct lock_struct); + data_len = br_lck->num_locks * sizeof(struct lock_struct); + + br_lck->lock_data = talloc_memdup(br_lck, data.dptr, data_len); + if (br_lck->lock_data == NULL) { + DEBUG(1, ("talloc_memdup failed\n")); + return false; + } + return true; +} + +/******************************************************************* + Fetch a set of byte range lock data from the database. + Leave the record locked. + TALLOC_FREE(brl) will release the lock in the destructor. +********************************************************************/ + +struct byte_range_lock *brl_get_locks(TALLOC_CTX *mem_ctx, files_struct *fsp) +{ + TDB_DATA key, data; + struct byte_range_lock *br_lck; + + br_lck = talloc_zero(mem_ctx, struct byte_range_lock); + if (br_lck == NULL) { + return NULL; + } + + br_lck->fsp = fsp; + + key.dptr = (uint8_t *)&fsp->file_id; + key.dsize = sizeof(struct file_id); + + br_lck->record = dbwrap_fetch_locked(brlock_db, br_lck, key); + + if (br_lck->record == NULL) { + DEBUG(3, ("Could not lock byte range lock entry\n")); + TALLOC_FREE(br_lck); + return NULL; + } + + data = dbwrap_record_get_value(br_lck->record); + + if (!brl_parse_data(br_lck, data)) { + TALLOC_FREE(br_lck); + return NULL; + } + + talloc_set_destructor(br_lck, byte_range_lock_destructor); + + if (DEBUGLEVEL >= 10) { + unsigned int i; + struct file_id_buf buf; + struct lock_struct *locks = br_lck->lock_data; + DBG_DEBUG("%u current locks on file_id %s\n", + br_lck->num_locks, + file_id_str_buf(fsp->file_id, &buf)); + for( i = 0; i < br_lck->num_locks; i++) { + print_lock_struct(i, &locks[i]); + } + } + + return br_lck; +} + +struct byte_range_lock *brl_get_locks_for_locking(TALLOC_CTX *mem_ctx, + files_struct *fsp, + TALLOC_CTX *req_mem_ctx, + const struct GUID *req_guid) +{ + struct byte_range_lock *br_lck = NULL; + + br_lck = brl_get_locks(mem_ctx, fsp); + if (br_lck == NULL) { + return NULL; + } + SMB_ASSERT(req_mem_ctx != NULL); + br_lck->req_mem_ctx = req_mem_ctx; + SMB_ASSERT(req_guid != NULL); + br_lck->req_guid = req_guid; + + return br_lck; +} + +struct brl_get_locks_readonly_state { + TALLOC_CTX *mem_ctx; + struct byte_range_lock **br_lock; +}; + +static void brl_get_locks_readonly_parser(TDB_DATA key, TDB_DATA data, + void *private_data) +{ + struct brl_get_locks_readonly_state *state = + (struct brl_get_locks_readonly_state *)private_data; + struct byte_range_lock *br_lck; + + br_lck = talloc_pooled_object( + state->mem_ctx, struct byte_range_lock, 1, data.dsize); + if (br_lck == NULL) { + *state->br_lock = NULL; + return; + } + *br_lck = (struct byte_range_lock) { 0 }; + if (!brl_parse_data(br_lck, data)) { + *state->br_lock = NULL; + return; + } + *state->br_lock = br_lck; +} + +struct byte_range_lock *brl_get_locks_readonly(files_struct *fsp) +{ + struct byte_range_lock *br_lock = NULL; + struct brl_get_locks_readonly_state state; + NTSTATUS status; + + DEBUG(10, ("seqnum=%d, fsp->brlock_seqnum=%d\n", + dbwrap_get_seqnum(brlock_db), fsp->brlock_seqnum)); + + if ((fsp->brlock_rec != NULL) + && (dbwrap_get_seqnum(brlock_db) == fsp->brlock_seqnum)) { + /* + * We have cached the brlock_rec and the database did not + * change. + */ + return fsp->brlock_rec; + } + + /* + * Parse the record fresh from the database + */ + + state.mem_ctx = fsp; + state.br_lock = &br_lock; + + status = dbwrap_parse_record( + brlock_db, + make_tdb_data((uint8_t *)&fsp->file_id, + sizeof(fsp->file_id)), + brl_get_locks_readonly_parser, &state); + + if (NT_STATUS_EQUAL(status,NT_STATUS_NOT_FOUND)) { + /* + * No locks on this file. Return an empty br_lock. + */ + br_lock = talloc_zero(fsp, struct byte_range_lock); + if (br_lock == NULL) { + return NULL; + } + + } else if (!NT_STATUS_IS_OK(status)) { + DEBUG(3, ("Could not parse byte range lock record: " + "%s\n", nt_errstr(status))); + return NULL; + } + if (br_lock == NULL) { + return NULL; + } + + br_lock->fsp = fsp; + br_lock->modified = false; + br_lock->record = NULL; + + /* + * Cache the brlock struct, invalidated when the dbwrap_seqnum + * changes. See beginning of this routine. + */ + TALLOC_FREE(fsp->brlock_rec); + fsp->brlock_rec = br_lock; + fsp->brlock_seqnum = dbwrap_get_seqnum(brlock_db); + + return br_lock; +} + +bool brl_cleanup_disconnected(struct file_id fid, uint64_t open_persistent_id) +{ + bool ret = false; + TALLOC_CTX *frame = talloc_stackframe(); + TDB_DATA key, val; + struct db_record *rec; + struct lock_struct *lock; + unsigned n, num; + struct file_id_buf buf; + NTSTATUS status; + + key = make_tdb_data((void*)&fid, sizeof(fid)); + + rec = dbwrap_fetch_locked(brlock_db, frame, key); + if (rec == NULL) { + DBG_INFO("failed to fetch record for file %s\n", + file_id_str_buf(fid, &buf)); + goto done; + } + + val = dbwrap_record_get_value(rec); + lock = (struct lock_struct*)val.dptr; + num = val.dsize / sizeof(struct lock_struct); + if (lock == NULL) { + DBG_DEBUG("no byte range locks for file %s\n", + file_id_str_buf(fid, &buf)); + ret = true; + goto done; + } + + for (n=0; n<num; n++) { + struct lock_context *ctx = &lock[n].context; + + if (!server_id_is_disconnected(&ctx->pid)) { + struct server_id_buf tmp; + DBG_INFO("byte range lock " + "%s used by server %s, do not cleanup\n", + file_id_str_buf(fid, &buf), + server_id_str_buf(ctx->pid, &tmp)); + goto done; + } + + if (ctx->smblctx != open_persistent_id) { + DBG_INFO("byte range lock %s expected smblctx %"PRIu64" " + "but found %"PRIu64", do not cleanup\n", + file_id_str_buf(fid, &buf), + open_persistent_id, + ctx->smblctx); + goto done; + } + } + + status = dbwrap_record_delete(rec); + if (!NT_STATUS_IS_OK(status)) { + DBG_INFO("failed to delete record " + "for file %s from %s, open %"PRIu64": %s\n", + file_id_str_buf(fid, &buf), + dbwrap_name(brlock_db), + open_persistent_id, + nt_errstr(status)); + goto done; + } + + DBG_DEBUG("file %s cleaned up %u entries from open %"PRIu64"\n", + file_id_str_buf(fid, &buf), + num, + open_persistent_id); + + ret = true; +done: + talloc_free(frame); + return ret; +} diff --git a/source3/locking/leases_db.c b/source3/locking/leases_db.c new file mode 100644 index 0000000..855d614 --- /dev/null +++ b/source3/locking/leases_db.c @@ -0,0 +1,726 @@ +/* + Unix SMB/CIFS implementation. + Map lease keys to file ids + Copyright (C) Volker Lendecke 2013 + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. + +*/ + +#include "includes.h" +#include "system/filesys.h" +#include "locking/leases_db.h" +#include "dbwrap/dbwrap.h" +#include "dbwrap/dbwrap_open.h" +#include "util_tdb.h" +#include "ndr.h" +#include "librpc/gen_ndr/ndr_leases_db.h" + +#undef DBGC_CLASS +#define DBGC_CLASS DBGC_LOCKING + +/* the leases database handle */ +static struct db_context *leases_db; + +bool leases_db_init(bool read_only) +{ + char *db_path; + + if (leases_db) { + return true; + } + + db_path = lock_path(talloc_tos(), "leases.tdb"); + if (db_path == NULL) { + return false; + } + + leases_db = db_open(NULL, db_path, 0, + TDB_DEFAULT| + TDB_VOLATILE| + TDB_CLEAR_IF_FIRST| + TDB_SEQNUM| + TDB_INCOMPATIBLE_HASH, + read_only ? O_RDONLY : O_RDWR|O_CREAT, 0644, + DBWRAP_LOCK_ORDER_4, DBWRAP_FLAG_NONE); + TALLOC_FREE(db_path); + if (leases_db == NULL) { + DEBUG(1, ("ERROR: Failed to initialise leases database\n")); + return false; + } + + return true; +} + +struct leases_db_key_buf { + uint8_t buf[32]; +}; + +static TDB_DATA leases_db_key(struct leases_db_key_buf *buf, + const struct GUID *client_guid, + const struct smb2_lease_key *lease_key) +{ + struct leases_db_key db_key = { + .client_guid = *client_guid, + .lease_key = *lease_key }; + DATA_BLOB blob = { .data = buf->buf, .length = sizeof(buf->buf) }; + enum ndr_err_code ndr_err; + + if (DEBUGLEVEL >= 10) { + DBG_DEBUG("\n"); + NDR_PRINT_DEBUG(leases_db_key, &db_key); + } + + ndr_err = ndr_push_struct_into_fixed_blob( + &blob, &db_key, (ndr_push_flags_fn_t)ndr_push_leases_db_key); + SMB_ASSERT(NDR_ERR_CODE_IS_SUCCESS(ndr_err)); + + return (TDB_DATA) { .dptr = buf->buf, .dsize = sizeof(buf->buf) }; +} + +struct leases_db_do_locked_state { + void (*fn)(struct leases_db_value *value, + bool *modified, + void *private_data); + void *private_data; + NTSTATUS status; +}; + +static void leases_db_do_locked_fn( + struct db_record *rec, + TDB_DATA db_value, + void *private_data) +{ + struct leases_db_do_locked_state *state = private_data; + DATA_BLOB blob = { .data = db_value.dptr, .length = db_value.dsize }; + struct leases_db_value *value = NULL; + enum ndr_err_code ndr_err; + bool modified = false; + + value = talloc_zero(talloc_tos(), struct leases_db_value); + if (value == NULL) { + state->status = NT_STATUS_NO_MEMORY; + goto done; + } + + if (blob.length != 0) { + ndr_err = ndr_pull_struct_blob_all( + &blob, + value, + value, + (ndr_pull_flags_fn_t)ndr_pull_leases_db_value); + if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) { + DBG_ERR("ndr_pull_struct_blob_failed: %s\n", + ndr_errstr(ndr_err)); + state->status = ndr_map_error2ntstatus(ndr_err); + goto done; + } + } + + state->fn(value, &modified, state->private_data); + + if (!modified) { + goto done; + } + + if (value->num_files == 0) { + state->status = dbwrap_record_delete(rec); + if (!NT_STATUS_IS_OK(state->status)) { + DBG_ERR("dbwrap_record_delete returned %s\n", + nt_errstr(state->status)); + } + goto done; + } + + ndr_err = ndr_push_struct_blob( + &blob, + value, + value, + (ndr_push_flags_fn_t)ndr_push_leases_db_value); + if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) { + DBG_ERR("ndr_push_struct_blob_failed: %s\n", + ndr_errstr(ndr_err)); + state->status = ndr_map_error2ntstatus(ndr_err); + goto done; + } + + if (DEBUGLEVEL >= 10) { + DBG_DEBUG("\n"); + NDR_PRINT_DEBUG(leases_db_value, value); + } + + db_value = make_tdb_data(blob.data, blob.length); + + state->status = dbwrap_record_store(rec, db_value, 0); + if (!NT_STATUS_IS_OK(state->status)) { + DBG_ERR("dbwrap_record_store returned %s\n", + nt_errstr(state->status)); + } + +done: + TALLOC_FREE(value); +} + +static NTSTATUS leases_db_do_locked( + const struct GUID *client_guid, + const struct smb2_lease_key *lease_key, + void (*fn)(struct leases_db_value *value, + bool *modified, + void *private_data), + void *private_data) +{ + struct leases_db_key_buf keybuf; + TDB_DATA db_key = leases_db_key(&keybuf, client_guid, lease_key); + struct leases_db_do_locked_state state = { + .fn = fn, .private_data = private_data, + }; + NTSTATUS status; + + if (!leases_db_init(false)) { + return NT_STATUS_INTERNAL_ERROR; + } + + status = dbwrap_do_locked( + leases_db, db_key, leases_db_do_locked_fn, &state); + if (!NT_STATUS_IS_OK(status)) { + return status; + } + return state.status; +} + +struct leases_db_add_state { + const struct file_id *id; + uint32_t current_state; + uint16_t lease_version; + uint16_t epoch; + const char *servicepath; + const char *base_name; + const char *stream_name; + NTSTATUS status; +}; + +static void leases_db_add_fn( + struct leases_db_value *value, bool *modified, void *private_data) +{ + struct leases_db_add_state *state = private_data; + struct leases_db_file *tmp = NULL; + uint32_t i; + + /* id must be unique. */ + for (i = 0; i < value->num_files; i++) { + if (file_id_equal(state->id, &value->files[i].id)) { + state->status = NT_STATUS_OBJECT_NAME_COLLISION; + return; + } + } + + if (value->num_files == 0) { + /* new record */ + value->current_state = state->current_state; + value->lease_version = state->lease_version; + value->epoch = state->epoch; + } + + tmp = talloc_realloc( + value, + value->files, + struct leases_db_file, + value->num_files + 1); + if (tmp == NULL) { + state->status = NT_STATUS_NO_MEMORY; + return; + } + value->files = tmp; + + value->files[value->num_files] = (struct leases_db_file) { + .id = *state->id, + .servicepath = state->servicepath, + .base_name = state->base_name, + .stream_name = state->stream_name, + }; + value->num_files += 1; + + *modified = true; +} + +NTSTATUS leases_db_add(const struct GUID *client_guid, + const struct smb2_lease_key *lease_key, + const struct file_id *id, + uint32_t current_state, + uint16_t lease_version, + uint16_t epoch, + const char *servicepath, + const char *base_name, + const char *stream_name) +{ + struct leases_db_add_state state = { + .id = id, + .current_state = current_state, + .lease_version = lease_version, + .epoch = epoch, + .servicepath = servicepath, + .base_name = base_name, + .stream_name = stream_name, + }; + NTSTATUS status; + + status = leases_db_do_locked( + client_guid, lease_key, leases_db_add_fn, &state); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("leases_db_do_locked failed: %s\n", + nt_errstr(status)); + return status; + } + return state.status; +} + +struct leases_db_del_state { + const struct file_id *id; + NTSTATUS status; +}; + +static void leases_db_del_fn( + struct leases_db_value *value, bool *modified, void *private_data) +{ + struct leases_db_del_state *state = private_data; + uint32_t i; + + for (i = 0; i < value->num_files; i++) { + if (file_id_equal(state->id, &value->files[i].id)) { + break; + } + } + if (i == value->num_files) { + state->status = NT_STATUS_NOT_FOUND; + return; + } + + value->files[i] = value->files[value->num_files-1]; + value->num_files -= 1; + + *modified = true; +} + +NTSTATUS leases_db_del(const struct GUID *client_guid, + const struct smb2_lease_key *lease_key, + const struct file_id *id) +{ + struct leases_db_del_state state = { .id = id }; + NTSTATUS status; + + status = leases_db_do_locked( + client_guid, lease_key, leases_db_del_fn, &state); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("leases_db_do_locked failed: %s\n", + nt_errstr(status)); + return status; + } + return state.status; +} + +struct leases_db_fetch_state { + void (*parser)(uint32_t num_files, + const struct leases_db_file *files, + void *private_data); + void *private_data; + NTSTATUS status; +}; + +static void leases_db_parser(TDB_DATA key, TDB_DATA data, void *private_data) +{ + struct leases_db_fetch_state *state = + (struct leases_db_fetch_state *)private_data; + DATA_BLOB blob = { .data = data.dptr, .length = data.dsize }; + enum ndr_err_code ndr_err; + struct leases_db_value *value; + + value = talloc(talloc_tos(), struct leases_db_value); + if (value == NULL) { + state->status = NT_STATUS_NO_MEMORY; + return; + } + + ndr_err = ndr_pull_struct_blob_all( + &blob, value, value, + (ndr_pull_flags_fn_t)ndr_pull_leases_db_value); + if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) { + DEBUG(10, ("%s: ndr_pull_struct_blob_failed: %s\n", + __func__, ndr_errstr(ndr_err))); + TALLOC_FREE(value); + state->status = ndr_map_error2ntstatus(ndr_err); + return; + } + + if (DEBUGLEVEL >= 10) { + DEBUG(10, ("%s:\n", __func__)); + NDR_PRINT_DEBUG(leases_db_value, value); + } + + state->parser(value->num_files, + value->files, + state->private_data); + + TALLOC_FREE(value); + state->status = NT_STATUS_OK; +} + +NTSTATUS leases_db_parse(const struct GUID *client_guid, + const struct smb2_lease_key *lease_key, + void (*parser)(uint32_t num_files, + const struct leases_db_file *files, + void *private_data), + void *private_data) +{ + struct leases_db_key_buf keybuf; + TDB_DATA db_key = leases_db_key(&keybuf, client_guid, lease_key); + struct leases_db_fetch_state state; + NTSTATUS status; + + if (!leases_db_init(true)) { + return NT_STATUS_INTERNAL_ERROR; + } + + state = (struct leases_db_fetch_state) { + .parser = parser, + .private_data = private_data, + .status = NT_STATUS_OK + }; + + status = dbwrap_parse_record(leases_db, db_key, leases_db_parser, + &state); + if (!NT_STATUS_IS_OK(status)) { + return status; + } + return state.status; +} + +struct leases_db_rename_state { + const struct file_id *id; + const char *servicename_new; + const char *filename_new; + const char *stream_name_new; + NTSTATUS status; +}; + +static void leases_db_rename_fn( + struct leases_db_value *value, bool *modified, void *private_data) +{ + struct leases_db_rename_state *state = private_data; + struct leases_db_file *file = NULL; + uint32_t i; + + /* id must exist. */ + for (i = 0; i < value->num_files; i++) { + if (file_id_equal(state->id, &value->files[i].id)) { + break; + } + } + if (i == value->num_files) { + state->status = NT_STATUS_NOT_FOUND; + return; + } + + file = &value->files[i]; + file->servicepath = state->servicename_new; + file->base_name = state->filename_new; + file->stream_name = state->stream_name_new; + + *modified = true; +} + +NTSTATUS leases_db_rename(const struct GUID *client_guid, + const struct smb2_lease_key *lease_key, + const struct file_id *id, + const char *servicename_new, + const char *filename_new, + const char *stream_name_new) +{ + struct leases_db_rename_state state = { + .id = id, + .servicename_new = servicename_new, + .filename_new = filename_new, + .stream_name_new = stream_name_new, + }; + NTSTATUS status; + + status = leases_db_do_locked( + client_guid, lease_key, leases_db_rename_fn, &state); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("leases_db_do_locked failed: %s\n", + nt_errstr(status)); + return status; + } + return state.status; +} + +struct leases_db_set_state { + uint32_t current_state; + bool breaking; + uint32_t breaking_to_requested; + uint32_t breaking_to_required; + uint16_t lease_version; + uint16_t epoch; +}; + +static void leases_db_set_fn( + struct leases_db_value *value, bool *modified, void *private_data) +{ + struct leases_db_set_state *state = private_data; + + if (value->num_files == 0) { + DBG_WARNING("leases_db_set on new entry\n"); + return; + } + value->current_state = state->current_state; + value->breaking = state->breaking; + value->breaking_to_requested = state->breaking_to_requested; + value->breaking_to_required = state->breaking_to_required; + value->lease_version = state->lease_version; + value->epoch = state->epoch; + *modified = true; +} + +NTSTATUS leases_db_set(const struct GUID *client_guid, + const struct smb2_lease_key *lease_key, + uint32_t current_state, + bool breaking, + uint32_t breaking_to_requested, + uint32_t breaking_to_required, + uint16_t lease_version, + uint16_t epoch) +{ + struct leases_db_set_state state = { + .current_state = current_state, + .breaking = breaking, + .breaking_to_requested = breaking_to_requested, + .breaking_to_required = breaking_to_required, + .lease_version = lease_version, + .epoch = epoch, + }; + NTSTATUS status; + + status = leases_db_do_locked( + client_guid, lease_key, leases_db_set_fn, &state); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("leases_db_do_locked failed: %s\n", + nt_errstr(status)); + return status; + } + return NT_STATUS_OK; +} + +struct leases_db_get_state { + const struct file_id *file_id; + uint32_t *current_state; + bool *breaking; + uint32_t *breaking_to_requested; + uint32_t *breaking_to_required; + uint16_t *lease_version; + uint16_t *epoch; + NTSTATUS status; +}; + +static void leases_db_get_fn(TDB_DATA key, TDB_DATA data, void *private_data) +{ + struct leases_db_get_state *state = private_data; + DATA_BLOB blob = { .data = data.dptr, .length = data.dsize }; + enum ndr_err_code ndr_err; + struct leases_db_value *value; + uint32_t i; + + value = talloc(talloc_tos(), struct leases_db_value); + if (value == NULL) { + state->status = NT_STATUS_NO_MEMORY; + return; + } + + ndr_err = ndr_pull_struct_blob_all( + &blob, value, value, + (ndr_pull_flags_fn_t)ndr_pull_leases_db_value); + if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) { + DBG_ERR("ndr_pull_struct_blob_failed: %s\n", + ndr_errstr(ndr_err)); + TALLOC_FREE(value); + state->status = ndr_map_error2ntstatus(ndr_err); + return; + } + + if (DEBUGLEVEL >= 10) { + DBG_DEBUG("\n"); + NDR_PRINT_DEBUG(leases_db_value, value); + } + + /* id must exist. */ + for (i = 0; i < value->num_files; i++) { + if (file_id_equal(state->file_id, &value->files[i].id)) { + break; + } + } + + if (i == value->num_files) { + state->status = NT_STATUS_NOT_FOUND; + TALLOC_FREE(value); + return; + } + + if (state->current_state != NULL) { + *state->current_state = value->current_state; + }; + if (state->breaking != NULL) { + *state->breaking = value->breaking; + }; + if (state->breaking_to_requested != NULL) { + *state->breaking_to_requested = value->breaking_to_requested; + }; + if (state->breaking_to_required != NULL) { + *state->breaking_to_required = value->breaking_to_required; + }; + if (state->lease_version != NULL) { + *state->lease_version = value->lease_version; + }; + if (state->epoch != NULL) { + *state->epoch = value->epoch; + }; + + TALLOC_FREE(value); + state->status = NT_STATUS_OK; +} + +NTSTATUS leases_db_get(const struct GUID *client_guid, + const struct smb2_lease_key *lease_key, + const struct file_id *file_id, + uint32_t *current_state, + bool *breaking, + uint32_t *breaking_to_requested, + uint32_t *breaking_to_required, + uint16_t *lease_version, + uint16_t *epoch) +{ + struct leases_db_get_state state = { + .file_id = file_id, + .current_state = current_state, + .breaking = breaking, + .breaking_to_requested = breaking_to_requested, + .breaking_to_required = breaking_to_required, + .lease_version = lease_version, + .epoch = epoch, + }; + struct leases_db_key_buf keybuf; + TDB_DATA db_key = leases_db_key(&keybuf, client_guid, lease_key); + NTSTATUS status; + + if (!leases_db_init(true)) { + return NT_STATUS_INTERNAL_ERROR; + } + + status = dbwrap_parse_record( + leases_db, db_key, leases_db_get_fn, &state); + if (!NT_STATUS_IS_OK(status)) { + return status; + } + return state.status; +} + +struct leases_db_get_current_state_state { + int seqnum; + uint32_t current_state; + NTSTATUS status; +}; + +/* + * This function is an optimization that + * relies on the fact that the + * smb2_lease_state current_state + * (which is a uint32_t size) + * from struct leases_db_value is the first + * entry in the ndr-encoded struct leases_db_value. + * Read it without having to ndr decode all + * the values in struct leases_db_value. + */ + +static void leases_db_get_current_state_fn( + TDB_DATA key, TDB_DATA data, void *private_data) +{ + struct leases_db_get_current_state_state *state = private_data; + struct ndr_pull ndr; + enum ndr_err_code ndr_err; + + if (data.dsize < sizeof(uint32_t)) { + state->status = NT_STATUS_INTERNAL_DB_CORRUPTION; + return; + } + + state->seqnum = dbwrap_get_seqnum(leases_db); + + ndr = (struct ndr_pull) { + .data = data.dptr, .data_size = data.dsize, + }; + ndr_err = ndr_pull_uint32(&ndr, NDR_SCALARS, &state->current_state); + if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) { + state->status = ndr_map_error2ntstatus(ndr_err); + } +} + +NTSTATUS leases_db_get_current_state( + const struct GUID *client_guid, + const struct smb2_lease_key *lease_key, + int *database_seqnum, + uint32_t *current_state) +{ + struct leases_db_get_current_state_state state = { 0 }; + struct leases_db_key_buf keybuf; + TDB_DATA db_key = { 0 }; + NTSTATUS status; + + if (!leases_db_init(true)) { + return NT_STATUS_INTERNAL_ERROR; + } + + state.seqnum = dbwrap_get_seqnum(leases_db); + if (*database_seqnum == state.seqnum) { + return NT_STATUS_OK; + } + + db_key = leases_db_key(&keybuf, client_guid, lease_key); + + status = dbwrap_parse_record( + leases_db, db_key, leases_db_get_current_state_fn, &state); + if (!NT_STATUS_IS_OK(status)) { + return status; + } + *database_seqnum = state.seqnum; + *current_state = state.current_state; + + return NT_STATUS_OK; +} + +NTSTATUS leases_db_copy_file_ids(TALLOC_CTX *mem_ctx, + uint32_t num_files, + const struct leases_db_file *files, + struct file_id **pp_ids) +{ + uint32_t i; + struct file_id *ids = talloc_array(mem_ctx, + struct file_id, + num_files); + if (ids == NULL) { + return NT_STATUS_NO_MEMORY; + } + + for (i = 0; i < num_files; i++) { + ids[i] = files[i].id; + } + *pp_ids = ids; + return NT_STATUS_OK; +} diff --git a/source3/locking/leases_db.h b/source3/locking/leases_db.h new file mode 100644 index 0000000..9c149c1 --- /dev/null +++ b/source3/locking/leases_db.h @@ -0,0 +1,80 @@ +/* + * Unix SMB/CIFS implementation. + * leases.tdb functions + * + * Copyright (C) Volker Lendecke 2014 + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 3 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program. If not, see <http://www.gnu.org/licenses/>. + */ + +#ifndef _LEASES_DB_H_ +#define _LEASES_DB_H_ + +struct GUID; +struct smb2_lease_key; +struct file_id; +struct leases_db_file; + +bool leases_db_init(bool read_only); +NTSTATUS leases_db_add(const struct GUID *client_guid, + const struct smb2_lease_key *lease_key, + const struct file_id *id, + uint32_t current_state, + uint16_t lease_version, + uint16_t epoch, + const char *servicepath, + const char *filename, + const char *stream_name); +NTSTATUS leases_db_del(const struct GUID *client_guid, + const struct smb2_lease_key *lease_key, + const struct file_id *id); +NTSTATUS leases_db_parse(const struct GUID *client_guid, + const struct smb2_lease_key *lease_key, + void (*parser)(uint32_t num_files, + const struct leases_db_file *files, + void *private_data), + void *private_data); +NTSTATUS leases_db_rename(const struct GUID *client_guid, + const struct smb2_lease_key *lease_key, + const struct file_id *id, + const char *servicepath_new, + const char *filename_new, + const char *stream_name_new); +NTSTATUS leases_db_set(const struct GUID *client_guid, + const struct smb2_lease_key *lease_key, + uint32_t current_state, + bool breaking, + uint32_t breaking_to_requested, + uint32_t breaking_to_required, + uint16_t lease_version, + uint16_t epoch); +NTSTATUS leases_db_get(const struct GUID *client_guid, + const struct smb2_lease_key *lease_key, + const struct file_id *file_id, + uint32_t *current_state, + bool *breaking, + uint32_t *breaking_to_requested, + uint32_t *breaking_to_required, + uint16_t *lease_version, + uint16_t *epoch); +NTSTATUS leases_db_get_current_state( + const struct GUID *client_guid, + const struct smb2_lease_key *lease_key, + int *database_seqnum, + uint32_t *current_state); +NTSTATUS leases_db_copy_file_ids(TALLOC_CTX *mem_ctx, + uint32_t num_files, + const struct leases_db_file *files, + struct file_id **pp_ids); +#endif /* _LEASES_DB_H_ */ diff --git a/source3/locking/leases_util.c b/source3/locking/leases_util.c new file mode 100644 index 0000000..9ae4081 --- /dev/null +++ b/source3/locking/leases_util.c @@ -0,0 +1,77 @@ +/* + Unix SMB/CIFS implementation. + Lease utility functions + + Copyright (C) Jeremy Allison 2017. + Copyright (C) Stefan (metze) Metzmacher 2017. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. +*/ + +#define DBGC_CLASS DBGC_LOCKING +#include "includes.h" +#include "../librpc/gen_ndr/open_files.h" +#include "locking/proto.h" +#include "smbd/globals.h" +#include "locking/leases_db.h" + +uint32_t map_oplock_to_lease_type(uint16_t op_type) +{ + uint32_t ret; + + switch(op_type) { + case BATCH_OPLOCK: + case BATCH_OPLOCK|EXCLUSIVE_OPLOCK: + ret = SMB2_LEASE_READ|SMB2_LEASE_WRITE|SMB2_LEASE_HANDLE; + break; + case EXCLUSIVE_OPLOCK: + ret = SMB2_LEASE_READ|SMB2_LEASE_WRITE; + break; + case LEVEL_II_OPLOCK: + ret = SMB2_LEASE_READ; + break; + default: + ret = SMB2_LEASE_NONE; + break; + } + return ret; +} + +uint32_t fsp_lease_type(struct files_struct *fsp) +{ + NTSTATUS status; + + if (fsp->oplock_type != LEASE_OPLOCK) { + uint32_t type = map_oplock_to_lease_type(fsp->oplock_type); + return type; + } + + status = leases_db_get_current_state( + fsp_client_guid(fsp), + &fsp->lease->lease.lease_key, + &fsp->leases_db_seqnum, + &fsp->lease_type); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("leases_db_get_current_state failed: %s\n", + nt_errstr(status)); + fsp->lease_type = 0; /* no lease */ + } + + return fsp->lease_type; +} + +const struct GUID *fsp_client_guid(const files_struct *fsp) +{ + return &fsp->conn->sconn->client->global->client_guid; +} diff --git a/source3/locking/locking.c b/source3/locking/locking.c new file mode 100644 index 0000000..befdc10 --- /dev/null +++ b/source3/locking/locking.c @@ -0,0 +1,1186 @@ +/* + Unix SMB/CIFS implementation. + Locking functions + Copyright (C) Andrew Tridgell 1992-2000 + Copyright (C) Jeremy Allison 1992-2006 + Copyright (C) Volker Lendecke 2005 + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. + + Revision History: + + 12 aug 96: Erik.Devriendt@te6.siemens.be + added support for shared memory implementation of share mode locking + + May 1997. Jeremy Allison (jallison@whistle.com). Modified share mode + locking to deal with multiple share modes per open file. + + September 1997. Jeremy Allison (jallison@whistle.com). Added oplock + support. + + rewritten completely to use new tdb code. Tridge, Dec '99 + + Added POSIX locking support. Jeremy Allison (jeremy@valinux.com), Apr. 2000. + Added Unix Extensions POSIX locking support. Jeremy Allison Mar 2006. +*/ + +#include "includes.h" +#include "lib/util/time_basic.h" +#include "system/filesys.h" +#include "lib/util/server_id.h" +#include "share_mode_lock.h" +#include "share_mode_lock_private.h" +#include "locking/proto.h" +#include "smbd/globals.h" +#include "dbwrap/dbwrap.h" +#include "dbwrap/dbwrap_open.h" +#include "../libcli/security/security.h" +#include "serverid.h" +#include "messages.h" +#include "util_tdb.h" +#include "../librpc/gen_ndr/ndr_open_files.h" +#include "librpc/gen_ndr/ndr_file_id.h" +#include "librpc/gen_ndr/ndr_leases_db.h" +#include "locking/leases_db.h" + +#undef DBGC_CLASS +#define DBGC_CLASS DBGC_LOCKING + +#define NO_LOCKING_COUNT (-1) + +/**************************************************************************** + Debugging aids :-). +****************************************************************************/ + +const char *lock_type_name(enum brl_type lock_type) +{ + switch (lock_type) { + case READ_LOCK: + return "READ"; + case WRITE_LOCK: + return "WRITE"; + default: + return "other"; + } +} + +const char *lock_flav_name(enum brl_flavour lock_flav) +{ + return (lock_flav == WINDOWS_LOCK) ? "WINDOWS_LOCK" : "POSIX_LOCK"; +} + +/**************************************************************************** + Utility function called to see if a file region is locked. + Called in the read/write codepath. +****************************************************************************/ + +void init_strict_lock_struct(files_struct *fsp, + uint64_t smblctx, + br_off start, + br_off size, + enum brl_type lock_type, + enum brl_flavour lock_flav, + struct lock_struct *plock) +{ + SMB_ASSERT(lock_type == READ_LOCK || lock_type == WRITE_LOCK); + + plock->context.smblctx = smblctx; + plock->context.tid = fsp->conn->cnum; + plock->context.pid = messaging_server_id(fsp->conn->sconn->msg_ctx); + plock->start = start; + plock->size = size; + plock->fnum = fsp->fnum; + plock->lock_type = lock_type; + plock->lock_flav = lp_posix_cifsu_locktype(fsp); +} + +bool strict_lock_check_default(files_struct *fsp, struct lock_struct *plock) +{ + struct byte_range_lock *br_lck; + int strict_locking = lp_strict_locking(fsp->conn->params); + bool ret = False; + + if (plock->size == 0) { + return True; + } + + if (!lp_locking(fsp->conn->params) || !strict_locking) { + return True; + } + + if (strict_locking == Auto) { + uint32_t lease_type = fsp_lease_type(fsp); + + if ((lease_type & SMB2_LEASE_READ) && + (plock->lock_type == READ_LOCK)) + { + DBG_DEBUG("optimisation - read lease on file %s\n", + fsp_str_dbg(fsp)); + return true; + } + + if ((lease_type & SMB2_LEASE_WRITE) && + (plock->lock_type == WRITE_LOCK)) + { + DBG_DEBUG("optimisation - write lease on file %s\n", + fsp_str_dbg(fsp)); + return true; + } + } + + br_lck = brl_get_locks_readonly(fsp); + if (!br_lck) { + return true; + } + ret = brl_locktest(br_lck, plock); + + if (!ret) { + /* + * We got a lock conflict. Retry with rw locks to enable + * autocleanup. This is the slow path anyway. + */ + br_lck = brl_get_locks(talloc_tos(), fsp); + if (br_lck == NULL) { + return true; + } + ret = brl_locktest(br_lck, plock); + TALLOC_FREE(br_lck); + } + + DEBUG(10, ("strict_lock_default: flavour = %s brl start=%ju " + "len=%ju %s for fnum %ju file %s\n", + lock_flav_name(plock->lock_flav), + (uintmax_t)plock->start, (uintmax_t)plock->size, + ret ? "unlocked" : "locked", + (uintmax_t)plock->fnum, fsp_str_dbg(fsp))); + + return ret; +} + +/**************************************************************************** + Find out if a lock could be granted - return who is blocking us if we can't. +****************************************************************************/ + +NTSTATUS query_lock(files_struct *fsp, + uint64_t *psmblctx, + uint64_t *pcount, + uint64_t *poffset, + enum brl_type *plock_type, + enum brl_flavour lock_flav) +{ + struct byte_range_lock *br_lck = NULL; + + if (!fsp->fsp_flags.can_lock) { + return fsp->fsp_flags.is_directory ? + NT_STATUS_INVALID_DEVICE_REQUEST : + NT_STATUS_INVALID_HANDLE; + } + + if (!lp_locking(fsp->conn->params)) { + return NT_STATUS_OK; + } + + br_lck = brl_get_locks_readonly(fsp); + if (!br_lck) { + return NT_STATUS_NO_MEMORY; + } + + return brl_lockquery(br_lck, + psmblctx, + messaging_server_id(fsp->conn->sconn->msg_ctx), + poffset, + pcount, + plock_type, + lock_flav); +} + +static void increment_current_lock_count(files_struct *fsp, + enum brl_flavour lock_flav) +{ + if (lock_flav == WINDOWS_LOCK && + fsp->current_lock_count != NO_LOCKING_COUNT) { + /* blocking ie. pending, locks also count here, + * as this is an efficiency counter to avoid checking + * the lock db. on close. JRA. */ + + fsp->current_lock_count++; + } else { + /* Notice that this has had a POSIX lock request. + * We can't count locks after this so forget them. + */ + fsp->current_lock_count = NO_LOCKING_COUNT; + } +} + +static void decrement_current_lock_count(files_struct *fsp, + enum brl_flavour lock_flav) +{ + if (lock_flav == WINDOWS_LOCK && + fsp->current_lock_count != NO_LOCKING_COUNT) { + SMB_ASSERT(fsp->current_lock_count > 0); + fsp->current_lock_count--; + } +} + +/**************************************************************************** + Utility function called by locking requests. +****************************************************************************/ + +struct do_lock_state { + struct files_struct *fsp; + TALLOC_CTX *req_mem_ctx; + const struct GUID *req_guid; + uint64_t smblctx; + uint64_t count; + uint64_t offset; + enum brl_type lock_type; + enum brl_flavour lock_flav; + + struct server_id blocker_pid; + uint64_t blocker_smblctx; + NTSTATUS status; +}; + +static void do_lock_fn( + const uint8_t *buf, + size_t buflen, + bool *modified_dependent, + void *private_data) +{ + struct do_lock_state *state = private_data; + struct byte_range_lock *br_lck = NULL; + + br_lck = brl_get_locks_for_locking(talloc_tos(), + state->fsp, + state->req_mem_ctx, + state->req_guid); + if (br_lck == NULL) { + state->status = NT_STATUS_NO_MEMORY; + return; + } + + state->status = brl_lock( + br_lck, + state->smblctx, + messaging_server_id(state->fsp->conn->sconn->msg_ctx), + state->offset, + state->count, + state->lock_type, + state->lock_flav, + &state->blocker_pid, + &state->blocker_smblctx); + + TALLOC_FREE(br_lck); +} + +NTSTATUS do_lock(files_struct *fsp, + TALLOC_CTX *req_mem_ctx, + const struct GUID *req_guid, + uint64_t smblctx, + uint64_t count, + uint64_t offset, + enum brl_type lock_type, + enum brl_flavour lock_flav, + struct server_id *pblocker_pid, + uint64_t *psmblctx) +{ + struct do_lock_state state = { + .fsp = fsp, + .req_mem_ctx = req_mem_ctx, + .req_guid = req_guid, + .smblctx = smblctx, + .count = count, + .offset = offset, + .lock_type = lock_type, + .lock_flav = lock_flav, + }; + NTSTATUS status; + + /* silently return ok on print files as we don't do locking there */ + if (fsp->print_file) { + return NT_STATUS_OK; + } + + if (!fsp->fsp_flags.can_lock) { + if (fsp->fsp_flags.is_directory) { + return NT_STATUS_INVALID_DEVICE_REQUEST; + } + return NT_STATUS_INVALID_HANDLE; + } + + if (!lp_locking(fsp->conn->params)) { + return NT_STATUS_OK; + } + + /* NOTE! 0 byte long ranges ARE allowed and should be stored */ + + DBG_DEBUG("lock flavour %s lock type %s start=%"PRIu64" len=%"PRIu64" " + "requested for %s file %s\n", + lock_flav_name(lock_flav), + lock_type_name(lock_type), + offset, + count, + fsp_fnum_dbg(fsp), + fsp_str_dbg(fsp)); + + status = share_mode_do_locked(fsp->file_id, do_lock_fn, &state); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("share_mode_do_locked returned %s\n", + nt_errstr(status)); + return status; + } + + if (psmblctx != NULL) { + *psmblctx = state.blocker_smblctx; + } + if (pblocker_pid != NULL) { + *pblocker_pid = state.blocker_pid; + } + + DBG_DEBUG("returning status=%s\n", nt_errstr(state.status)); + + increment_current_lock_count(fsp, lock_flav); + + return state.status; +} + +/**************************************************************************** + Utility function called by unlocking requests. +****************************************************************************/ + +NTSTATUS do_unlock(files_struct *fsp, + uint64_t smblctx, + uint64_t count, + uint64_t offset, + enum brl_flavour lock_flav) +{ + bool ok = False; + struct byte_range_lock *br_lck = NULL; + + if (!fsp->fsp_flags.can_lock) { + return fsp->fsp_flags.is_directory ? + NT_STATUS_INVALID_DEVICE_REQUEST : + NT_STATUS_INVALID_HANDLE; + } + + if (!lp_locking(fsp->conn->params)) { + return NT_STATUS_OK; + } + + DBG_DEBUG("unlock start=%"PRIu64" len=%"PRIu64" requested for %s file " + "%s\n", + offset, + count, + fsp_fnum_dbg(fsp), + fsp_str_dbg(fsp)); + + br_lck = brl_get_locks(talloc_tos(), fsp); + if (!br_lck) { + return NT_STATUS_NO_MEMORY; + } + + ok = brl_unlock(br_lck, + smblctx, + messaging_server_id(fsp->conn->sconn->msg_ctx), + offset, + count, + lock_flav); + + TALLOC_FREE(br_lck); + + if (!ok) { + DEBUG(10,("do_unlock: returning ERRlock.\n" )); + return NT_STATUS_RANGE_NOT_LOCKED; + } + + decrement_current_lock_count(fsp, lock_flav); + return NT_STATUS_OK; +} + +/**************************************************************************** + Remove any locks on this fd. Called from file_close(). +****************************************************************************/ + +void locking_close_file(files_struct *fsp, + enum file_close_type close_type) +{ + struct byte_range_lock *br_lck; + + if (!lp_locking(fsp->conn->params)) { + return; + } + + /* If we have no outstanding locks or pending + * locks then we don't need to look in the lock db. + */ + + if (fsp->current_lock_count == 0) { + return; + } + + br_lck = brl_get_locks(talloc_tos(),fsp); + + if (br_lck) { + /* + * Unlocks must trigger dbwrap_watch watchers, + * normally in smbd_do_unlocking. Here it's done + * implictly, we're closing the file and thus remove a + * share mode. This will wake the waiters. + */ + brl_close_fnum(br_lck); + TALLOC_FREE(br_lck); + } +} + +/******************************************************************* + Print out a share mode. +********************************************************************/ + +char *share_mode_str(TALLOC_CTX *ctx, int num, + const struct file_id *id, + const struct share_mode_entry *e) +{ + struct server_id_buf tmp; + struct file_id_buf ftmp; + + return talloc_asprintf(ctx, "share_mode_entry[%d]: " + "pid = %s, share_access = 0x%x, private_options = 0x%x, " + "access_mask = 0x%x, mid = 0x%llx, type= 0x%x, gen_id = %llu, " + "uid = %u, flags = %u, file_id %s, name_hash = 0x%x", + num, + server_id_str_buf(e->pid, &tmp), + e->share_access, e->private_options, + e->access_mask, (unsigned long long)e->op_mid, + e->op_type, (unsigned long long)e->share_file_id, + (unsigned int)e->uid, (unsigned int)e->flags, + file_id_str_buf(*id, &ftmp), + (unsigned int)e->name_hash); +} + +struct rename_share_filename_state { + struct share_mode_lock *lck; + struct messaging_context *msg_ctx; + struct server_id self; + uint32_t orig_name_hash; + uint32_t new_name_hash; + struct file_rename_message msg; +}; + +static bool rename_lease_fn(struct share_mode_entry *e, + void *private_data) +{ + struct rename_share_filename_state *state = private_data; + struct share_mode_data *d = state->lck->data; + NTSTATUS status; + + status = leases_db_rename(&e->client_guid, + &e->lease_key, + &d->id, + d->servicepath, + d->base_name, + d->stream_name); + + if (!NT_STATUS_IS_OK(status)) { + /* Any error recovery possible here ? */ + DBG_WARNING("Failed to rename lease key for " + "renamed file %s:%s. %s\n", + d->base_name, + d->stream_name, + nt_errstr(status)); + } + + return false; +} + +/******************************************************************* + Sets the service name and filename for rename. + At this point we emit "file renamed" messages to all + process id's that have this file open. + Based on an initial code idea from SATOH Fumiyasu <fumiya@samba.gr.jp> +********************************************************************/ + +static bool rename_share_filename_fn( + struct share_mode_entry *e, + bool *modified, + void *private_data) +{ + struct rename_share_filename_state *state = private_data; + DATA_BLOB blob; + enum ndr_err_code ndr_err; + bool ok; + + /* + * If this is a hardlink to the inode with a different name, + * skip this. + */ + if (e->name_hash != state->orig_name_hash) { + return false; + } + e->name_hash = state->new_name_hash; + *modified = true; + + ok = server_id_equal(&e->pid, &state->self); + if (ok) { + return false; + } + + state->msg.share_file_id = e->share_file_id; + + ndr_err = ndr_push_struct_blob( + &blob, + talloc_tos(), + &state->msg, + (ndr_push_flags_fn_t)ndr_push_file_rename_message); + if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) { + DBG_DEBUG("ndr_push_file_rename_message failed: %s\n", + ndr_errstr(ndr_err)); + return false; + } + if (DEBUGLEVEL >= 10) { + struct server_id_buf tmp; + DBG_DEBUG("sending rename message to %s\n", + server_id_str_buf(e->pid, &tmp)); + NDR_PRINT_DEBUG(file_rename_message, &state->msg); + } + + messaging_send(state->msg_ctx, e->pid, MSG_SMB_FILE_RENAME, &blob); + + TALLOC_FREE(blob.data); + + return false; +} + +bool rename_share_filename(struct messaging_context *msg_ctx, + struct share_mode_lock *lck, + struct file_id id, + const char *servicepath, + uint32_t orig_name_hash, + uint32_t new_name_hash, + const struct smb_filename *smb_fname_dst) +{ + struct rename_share_filename_state state = { + .lck = lck, + .msg_ctx = msg_ctx, + .self = messaging_server_id(msg_ctx), + .orig_name_hash = orig_name_hash, + .new_name_hash = new_name_hash, + .msg.id = id, + .msg.servicepath = servicepath, + .msg.base_name = smb_fname_dst->base_name, + .msg.stream_name = smb_fname_dst->stream_name, + }; + struct share_mode_data *d = lck->data; + bool ok; + + DEBUG(10, ("rename_share_filename: servicepath %s newname %s\n", + servicepath, smb_fname_dst->base_name)); + + /* + * rename_internal_fsp() and rename_internals() add './' to + * head of newname if newname does not contain a '/'. + */ + + if (strncmp(state.msg.base_name, "./", 2) == 0) { + state.msg.base_name += 2; + } + + d->servicepath = talloc_strdup(d, state.msg.servicepath); + d->base_name = talloc_strdup(d, state.msg.base_name); + d->stream_name = talloc_strdup(d, state.msg.stream_name); + if ((d->servicepath == NULL) || + (d->base_name == NULL) || + ((state.msg.stream_name != NULL) && (d->stream_name == NULL))) { + DBG_WARNING("talloc failed\n"); + return false; + } + d->modified = True; + + ok = share_mode_forall_entries( + lck, rename_share_filename_fn, &state); + if (!ok) { + DBG_WARNING("share_mode_forall_entries failed\n"); + } + + ok = share_mode_forall_leases(lck, rename_lease_fn, &state); + if (!ok) { + /* + * Ignore error here. Not sure what to do.. + */ + DBG_WARNING("share_mode_forall_leases failed\n"); + } + + return True; +} + +void get_file_infos(struct file_id id, + uint32_t name_hash, + bool *delete_on_close, + struct timespec *write_time) +{ + struct share_mode_lock *lck; + + if (delete_on_close) { + *delete_on_close = false; + } + + if (write_time) { + *write_time = make_omit_timespec(); + } + + if (!(lck = fetch_share_mode_unlocked(talloc_tos(), id))) { + return; + } + + if (delete_on_close) { + *delete_on_close = is_delete_on_close_set(lck, name_hash); + } + + if (write_time) { + *write_time = get_share_mode_write_time(lck); + } + + TALLOC_FREE(lck); +} + +bool is_valid_share_mode_entry(const struct share_mode_entry *e) +{ + int num_props = 0; + + if (e->stale) { + return false; + } + + num_props += ((e->op_type == NO_OPLOCK) ? 1 : 0); + num_props += (EXCLUSIVE_OPLOCK_TYPE(e->op_type) ? 1 : 0); + num_props += (LEVEL_II_OPLOCK_TYPE(e->op_type) ? 1 : 0); + num_props += (e->op_type == LEASE_OPLOCK); + + if ((num_props > 1) && serverid_exists(&e->pid)) { + smb_panic("Invalid share mode entry"); + } + return (num_props != 0); +} + +struct find_lease_ref_state { + const struct GUID *client_guid; + const struct smb2_lease_key *lease_key; + bool found_same; +}; + +static bool find_lease_ref_fn( + struct share_mode_entry *e, + bool *modified, + void *private_data) +{ + struct find_lease_ref_state *state = private_data; + + if (e->stale) { + return false; + } + if (e->op_type != LEASE_OPLOCK) { + return false; + } + + state->found_same = smb2_lease_equal( + &e->client_guid, + &e->lease_key, + state->client_guid, + state->lease_key); + /* + * If we found a lease reference, look no further (i.e. return true) + */ + return state->found_same; +} + +NTSTATUS remove_lease_if_stale(struct share_mode_lock *lck, + const struct GUID *client_guid, + const struct smb2_lease_key *lease_key) +{ + struct find_lease_ref_state state = { + .client_guid = client_guid, .lease_key = lease_key, + }; + struct share_mode_data *d = lck->data; + NTSTATUS status; + bool ok; + + ok = share_mode_forall_entries(lck, find_lease_ref_fn, &state); + if (!ok) { + return NT_STATUS_INTERNAL_ERROR; + } + + if (state.found_same) { + return NT_STATUS_RESOURCE_IN_USE; + } + + status = leases_db_del(client_guid, lease_key, &d->id); + if (!NT_STATUS_IS_OK(status)) { + int level = DBGLVL_DEBUG; + + if (!NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND)) { + level = DBGLVL_ERR; + } + DBG_PREFIX(level, ("leases_db_del failed: %s\n", + nt_errstr(status))); + } + return status; +} + +bool share_entry_stale_pid(struct share_mode_entry *e) +{ + struct server_id_buf buf; + bool exists; + + if (e->stale) { + return true; + } + + exists = serverid_exists(&e->pid); + if (exists) { + DBG_DEBUG("PID %s still exists\n", + server_id_str_buf(e->pid, &buf)); + return false; + } + + DBG_DEBUG("PID %s does not exist anymore\n", + server_id_str_buf(e->pid, &buf)); + + e->stale = true; + + return true; +} + +/**************************************************************************** + Adds a delete on close token. +****************************************************************************/ + +static bool add_delete_on_close_token(struct share_mode_data *d, + uint32_t name_hash, + const struct security_token *nt_tok, + const struct security_unix_token *tok) +{ + struct delete_token *tmp, *dtl; + + tmp = talloc_realloc(d, d->delete_tokens, struct delete_token, + d->num_delete_tokens+1); + if (tmp == NULL) { + return false; + } + d->delete_tokens = tmp; + dtl = &d->delete_tokens[d->num_delete_tokens]; + + dtl->name_hash = name_hash; + dtl->delete_nt_token = dup_nt_token(d->delete_tokens, nt_tok); + if (dtl->delete_nt_token == NULL) { + return false; + } + dtl->delete_token = copy_unix_token(d->delete_tokens, tok); + if (dtl->delete_token == NULL) { + return false; + } + d->num_delete_tokens += 1; + d->modified = true; + return true; +} + +void reset_delete_on_close_lck(files_struct *fsp, + struct share_mode_lock *lck) +{ + struct share_mode_data *d = lck->data; + uint32_t i; + + for (i=0; i<d->num_delete_tokens; i++) { + struct delete_token *dt = &d->delete_tokens[i]; + + if (dt->name_hash == fsp->name_hash) { + d->modified = true; + + /* Delete this entry. */ + TALLOC_FREE(dt->delete_nt_token); + TALLOC_FREE(dt->delete_token); + *dt = d->delete_tokens[d->num_delete_tokens-1]; + d->num_delete_tokens -= 1; + } + } +} + +struct set_delete_on_close_state { + struct messaging_context *msg_ctx; + DATA_BLOB blob; +}; + +static bool set_delete_on_close_fn( + struct share_mode_entry *e, + bool *modified, + void *private_data) +{ + struct set_delete_on_close_state *state = private_data; + NTSTATUS status; + + status = messaging_send( + state->msg_ctx, + e->pid, + MSG_SMB_NOTIFY_CANCEL_DELETED, + &state->blob); + + if (!NT_STATUS_IS_OK(status)) { + struct server_id_buf tmp; + DBG_DEBUG("messaging_send to %s returned %s\n", + server_id_str_buf(e->pid, &tmp), + nt_errstr(status)); + } + + return false; +} + +/**************************************************************************** + Sets the delete on close flag over all share modes on this file. + Modify the share mode entry for all files open + on this device and inode to tell other smbds we have + changed the delete on close flag. This will be noticed + in the close code, the last closer will delete the file + if flag is set. + This makes a copy of any struct security_unix_token into the + lck entry. This function is used when the lock is already granted. +****************************************************************************/ + +void set_delete_on_close_lck(files_struct *fsp, + struct share_mode_lock *lck, + const struct security_token *nt_tok, + const struct security_unix_token *tok) +{ + struct share_mode_data *d = lck->data; + struct set_delete_on_close_state state = { + .msg_ctx = fsp->conn->sconn->msg_ctx + }; + uint32_t i; + bool ret; + enum ndr_err_code ndr_err; + + SMB_ASSERT(nt_tok != NULL); + SMB_ASSERT(tok != NULL); + + for (i=0; i<d->num_delete_tokens; i++) { + struct delete_token *dt = &d->delete_tokens[i]; + if (dt->name_hash == fsp->name_hash) { + d->modified = true; + + /* Replace this token with the given tok. */ + TALLOC_FREE(dt->delete_nt_token); + dt->delete_nt_token = dup_nt_token(dt, nt_tok); + SMB_ASSERT(dt->delete_nt_token != NULL); + TALLOC_FREE(dt->delete_token); + dt->delete_token = copy_unix_token(dt, tok); + SMB_ASSERT(dt->delete_token != NULL); + + return; + } + } + + ret = add_delete_on_close_token(lck->data, fsp->name_hash, nt_tok, tok); + SMB_ASSERT(ret); + + ndr_err = ndr_push_struct_blob( + &state.blob, + talloc_tos(), + &fsp->file_id, + (ndr_push_flags_fn_t)ndr_push_file_id); + if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) { + DEBUG(10, ("ndr_push_file_id failed: %s\n", + ndr_errstr(ndr_err))); + } + + ret = share_mode_forall_entries( + lck, set_delete_on_close_fn, &state); + if (!ret) { + DBG_DEBUG("share_mode_forall_entries failed\n"); + } + + TALLOC_FREE(state.blob.data); +} + +bool set_delete_on_close(files_struct *fsp, bool delete_on_close, + const struct security_token *nt_tok, + const struct security_unix_token *tok) +{ + struct share_mode_lock *lck; + + DEBUG(10,("set_delete_on_close: %s delete on close flag for " + "%s, file %s\n", + delete_on_close ? "Adding" : "Removing", fsp_fnum_dbg(fsp), + fsp_str_dbg(fsp))); + + lck = get_existing_share_mode_lock(talloc_tos(), fsp->file_id); + if (lck == NULL) { + return False; + } + + if (delete_on_close) { + set_delete_on_close_lck(fsp, lck, nt_tok, tok); + } else { + reset_delete_on_close_lck(fsp, lck); + } + + if (fsp->fsp_flags.is_directory) { + SMB_ASSERT(!is_ntfs_stream_smb_fname(fsp->fsp_name)); + send_stat_cache_delete_message(fsp->conn->sconn->msg_ctx, + fsp->fsp_name->base_name); + } + + TALLOC_FREE(lck); + + fsp->fsp_flags.delete_on_close = delete_on_close; + + return True; +} + +static struct delete_token *find_delete_on_close_token( + struct share_mode_data *d, uint32_t name_hash) +{ + uint32_t i; + + DBG_DEBUG("name_hash = 0x%"PRIx32"\n", name_hash); + + for (i=0; i<d->num_delete_tokens; i++) { + struct delete_token *dt = &d->delete_tokens[i]; + + DBG_DEBUG("dt->name_hash = 0x%"PRIx32"\n", + dt->name_hash); + if (dt->name_hash == name_hash) { + return dt; + } + } + return NULL; +} + +/**************************************************************************** + Return the NT token and UNIX token if there's a match. Return true if + found, false if not. +****************************************************************************/ + +bool get_delete_on_close_token(struct share_mode_lock *lck, + uint32_t name_hash, + const struct security_token **pp_nt_tok, + const struct security_unix_token **pp_tok) +{ + struct delete_token *dt; + + dt = find_delete_on_close_token(lck->data, name_hash); + if (dt == NULL) { + return false; + } + *pp_nt_tok = dt->delete_nt_token; + *pp_tok = dt->delete_token; + return true; +} + +bool is_delete_on_close_set(struct share_mode_lock *lck, uint32_t name_hash) +{ + return find_delete_on_close_token(lck->data, name_hash) != NULL; +} + +bool set_sticky_write_time(struct file_id fileid, struct timespec write_time) +{ + struct share_mode_lock *lck; + struct file_id_buf ftmp; + struct timeval_buf tbuf; + NTTIME nt = full_timespec_to_nt_time(&write_time); + + DBG_INFO("%s id=%s\n", + timespec_string_buf(&write_time, true, &tbuf), + file_id_str_buf(fileid, &ftmp)); + + lck = get_existing_share_mode_lock(talloc_tos(), fileid); + if (lck == NULL) { + return False; + } + + if (lck->data->changed_write_time != nt) { + lck->data->modified = True; + lck->data->changed_write_time = nt; + } + + TALLOC_FREE(lck); + return True; +} + +bool set_write_time(struct file_id fileid, struct timespec write_time) +{ + struct share_mode_lock *lck; + struct file_id_buf idbuf; + struct timeval_buf tbuf; + NTTIME nt = full_timespec_to_nt_time(&write_time); + + DBG_INFO("%s id=%s\n", + timespec_string_buf(&write_time, true, &tbuf), + file_id_str_buf(fileid, &idbuf)); + + lck = get_existing_share_mode_lock(talloc_tos(), fileid); + if (lck == NULL) { + return False; + } + + if (lck->data->old_write_time != nt) { + lck->data->modified = True; + lck->data->old_write_time = nt; + } + + TALLOC_FREE(lck); + return True; +} + +struct timespec get_share_mode_write_time(struct share_mode_lock *lck) +{ + struct share_mode_data *d = lck->data; + + if (!null_nttime(d->changed_write_time)) { + return nt_time_to_full_timespec(d->changed_write_time); + } + return nt_time_to_full_timespec(d->old_write_time); +} + +struct file_has_open_streams_state { + bool found_one; +}; + +static bool file_has_open_streams_fn( + struct share_mode_entry *e, + bool *modified, + void *private_data) +{ + struct file_has_open_streams_state *state = private_data; + + if ((e->private_options & + NTCREATEX_FLAG_STREAM_BASEOPEN) == 0) { + return false; + } + + if (share_entry_stale_pid(e)) { + return false; + } + + state->found_one = true; + return true; +} + +bool file_has_open_streams(files_struct *fsp) +{ + struct file_has_open_streams_state state = { .found_one = false }; + struct share_mode_lock *lock = NULL; + bool ok; + + lock = get_existing_share_mode_lock(talloc_tos(), fsp->file_id); + if (lock == NULL) { + return false; + } + + ok = share_mode_forall_entries( + lock, file_has_open_streams_fn, &state); + TALLOC_FREE(lock); + + if (!ok) { + DBG_DEBUG("share_mode_forall_entries failed\n"); + return false; + } + return state.found_one; +} + +/* + * Walk share mode entries, looking at every lease only once + */ + +struct share_mode_forall_leases_state { + TALLOC_CTX *mem_ctx; + struct leases_db_key *leases; + bool (*fn)(struct share_mode_entry *e, + void *private_data); + void *private_data; + NTSTATUS status; +}; + +static bool share_mode_forall_leases_fn( + struct share_mode_entry *e, + bool *modified, + void *private_data) +{ + struct share_mode_forall_leases_state *state = private_data; + struct leases_db_key *leases = state->leases; + size_t i, num_leases; + bool stop; + + if (e->op_type != LEASE_OPLOCK) { + return false; + } + + num_leases = talloc_array_length(leases); + + for (i=0; i<num_leases; i++) { + struct leases_db_key *l = &leases[i]; + bool same = smb2_lease_equal( + &e->client_guid, + &e->lease_key, + &l->client_guid, + &l->lease_key); + if (same) { + return false; + } + } + + leases = talloc_realloc( + state->mem_ctx, + leases, + struct leases_db_key, + num_leases+1); + if (leases == NULL) { + state->status = NT_STATUS_NO_MEMORY; + return true; + } + leases[num_leases] = (struct leases_db_key) { + .client_guid = e->client_guid, + .lease_key = e->lease_key, + }; + state->leases = leases; + + stop = state->fn(e, state->private_data); + return stop; +} + +bool share_mode_forall_leases( + struct share_mode_lock *lck, + bool (*fn)(struct share_mode_entry *e, + void *private_data), + void *private_data) +{ + struct share_mode_forall_leases_state state = { + .mem_ctx = talloc_tos(), + .fn = fn, + .private_data = private_data + }; + bool ok; + + ok = share_mode_forall_entries( + lck, share_mode_forall_leases_fn, &state); + TALLOC_FREE(state.leases); + if (!ok) { + DBG_DEBUG("share_mode_forall_entries failed\n"); + return false; + } + + if (!NT_STATUS_IS_OK(state.status)) { + DBG_DEBUG("share_mode_forall_leases_fn returned %s\n", + nt_errstr(state.status)); + return false; + } + + return true; +} diff --git a/source3/locking/posix.c b/source3/locking/posix.c new file mode 100644 index 0000000..536d8bf --- /dev/null +++ b/source3/locking/posix.c @@ -0,0 +1,1374 @@ +/* + Unix SMB/CIFS implementation. + Locking functions + Copyright (C) Jeremy Allison 1992-2006 + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. + + Revision History: + + POSIX locking support. Jeremy Allison (jeremy@valinux.com), Apr. 2000. +*/ + +#include "includes.h" +#include "system/filesys.h" +#include "lib/util/server_id.h" +#include "locking/proto.h" +#include "dbwrap/dbwrap.h" +#include "dbwrap/dbwrap_rbt.h" +#include "util_tdb.h" +#include "smbd/fd_handle.h" + +#undef DBGC_CLASS +#define DBGC_CLASS DBGC_LOCKING + +/* + * The pending close database handle. + */ + +static struct db_context *posix_pending_close_db; + +/**************************************************************************** + First - the functions that deal with the underlying system locks - these + functions are used no matter if we're mapping CIFS Windows locks or CIFS + POSIX locks onto POSIX. +****************************************************************************/ + +/**************************************************************************** + Utility function to map a lock type correctly depending on the open + mode of a file. +****************************************************************************/ + +static int map_posix_lock_type( files_struct *fsp, enum brl_type lock_type) +{ + if ((lock_type == WRITE_LOCK) && !fsp->fsp_flags.can_write) { + /* + * Many UNIX's cannot get a write lock on a file opened read-only. + * Win32 locking semantics allow this. + * Do the best we can and attempt a read-only lock. + */ + DEBUG(10,("map_posix_lock_type: Downgrading write lock to read due to read-only file.\n")); + return F_RDLCK; + } + + /* + * This return should be the most normal, as we attempt + * to always open files read/write. + */ + + return (lock_type == READ_LOCK) ? F_RDLCK : F_WRLCK; +} + +/**************************************************************************** + Debugging aid :-). +****************************************************************************/ + +static const char *posix_lock_type_name(int lock_type) +{ + return (lock_type == F_RDLCK) ? "READ" : "WRITE"; +} + +/**************************************************************************** + Check to see if the given unsigned lock range is within the possible POSIX + range. Modifies the given args to be in range if possible, just returns + False if not. +****************************************************************************/ + +#define SMB_OFF_T_BITS (sizeof(off_t)*8) + +static bool posix_lock_in_range(off_t *offset_out, off_t *count_out, + uint64_t u_offset, uint64_t u_count) +{ + off_t offset = (off_t)u_offset; + off_t count = (off_t)u_count; + + /* + * For the type of system we are, attempt to + * find the maximum positive lock offset as an off_t. + */ + +#if defined(MAX_POSITIVE_LOCK_OFFSET) /* Some systems have arbitrary limits. */ + + off_t max_positive_lock_offset = (MAX_POSITIVE_LOCK_OFFSET); +#else + /* + * In this case off_t is 64 bits, + * and the underlying system can handle 64 bit signed locks. + */ + + off_t mask2 = ((off_t)0x4) << (SMB_OFF_T_BITS-4); + off_t mask = (mask2<<1); + off_t max_positive_lock_offset = ~mask; + +#endif + /* + * POSIX locks of length zero mean lock to end-of-file. + * Win32 locks of length zero are point probes. Ignore + * any Win32 locks of length zero. JRA. + */ + + if (count == 0) { + DEBUG(10,("posix_lock_in_range: count = 0, ignoring.\n")); + return False; + } + + /* + * If the given offset was > max_positive_lock_offset then we cannot map this at all + * ignore this lock. + */ + + if (u_offset & ~((uint64_t)max_positive_lock_offset)) { + DEBUG(10, ("posix_lock_in_range: (offset = %ju) offset > %ju " + "and we cannot handle this. Ignoring lock.\n", + (uintmax_t)u_offset, + (uintmax_t)max_positive_lock_offset)); + return False; + } + + /* + * We must truncate the count to less than max_positive_lock_offset. + */ + + if (u_count & ~((uint64_t)max_positive_lock_offset)) { + count = max_positive_lock_offset; + } + + /* + * Truncate count to end at max lock offset. + */ + + if (offset > INT64_MAX - count || + offset + count > max_positive_lock_offset) { + count = max_positive_lock_offset - offset; + } + + /* + * If we ate all the count, ignore this lock. + */ + + if (count == 0) { + DEBUG(10, ("posix_lock_in_range: Count = 0. Ignoring lock " + "u_offset = %ju, u_count = %ju\n", + (uintmax_t)u_offset, + (uintmax_t)u_count)); + return False; + } + + /* + * The mapping was successful. + */ + + DEBUG(10, ("posix_lock_in_range: offset_out = %ju, " + "count_out = %ju\n", + (uintmax_t)offset, (uintmax_t)count)); + + *offset_out = offset; + *count_out = count; + + return True; +} + +bool smb_vfs_call_lock(struct vfs_handle_struct *handle, + struct files_struct *fsp, int op, off_t offset, + off_t count, int type) +{ + VFS_FIND(lock); + return handle->fns->lock_fn(handle, fsp, op, offset, count, type); +} + +/**************************************************************************** + Actual function that does POSIX locks. Copes with 64 -> 32 bit cruft and + broken NFS implementations. +****************************************************************************/ + +static bool posix_fcntl_lock(files_struct *fsp, int op, off_t offset, off_t count, int type) +{ + bool ret; + + DEBUG(8,("posix_fcntl_lock %d %d %jd %jd %d\n", + fsp_get_io_fd(fsp),op,(intmax_t)offset,(intmax_t)count,type)); + + ret = SMB_VFS_LOCK(fsp, op, offset, count, type); + + if (!ret && ((errno == EFBIG) || (errno == ENOLCK) || (errno == EINVAL))) { + + if ((errno == EINVAL) && + (op != F_GETLK && + op != F_SETLK && + op != F_SETLKW)) { + DEBUG(0,("WARNING: OFD locks in use and no kernel " + "support. Try setting " + "'smbd:force process locks = true' " + "in smb.conf\n")); + } else { + DEBUG(0, ("WARNING: lock request at offset " + "%ju, length %ju returned\n", + (uintmax_t)offset, (uintmax_t)count)); + DEBUGADD(0, ("an %s error. This can happen when using 64 bit " + "lock offsets\n", strerror(errno))); + DEBUGADD(0, ("on 32 bit NFS mounted file systems.\n")); + } + + /* + * If the offset is > 0x7FFFFFFF then this will cause problems on + * 32 bit NFS mounted filesystems. Just ignore it. + */ + + if (offset & ~((off_t)0x7fffffff)) { + DEBUG(0,("Offset greater than 31 bits. Returning success.\n")); + return True; + } + + if (count & ~((off_t)0x7fffffff)) { + /* 32 bit NFS file system, retry with smaller offset */ + DEBUG(0,("Count greater than 31 bits - retrying with 31 bit truncated length.\n")); + errno = 0; + count &= 0x7fffffff; + ret = SMB_VFS_LOCK(fsp, op, offset, count, type); + } + } + + DEBUG(8,("posix_fcntl_lock: Lock call %s\n", ret ? "successful" : "failed")); + return ret; +} + +bool smb_vfs_call_getlock(struct vfs_handle_struct *handle, + struct files_struct *fsp, off_t *poffset, + off_t *pcount, int *ptype, pid_t *ppid) +{ + VFS_FIND(getlock); + return handle->fns->getlock_fn(handle, fsp, poffset, pcount, ptype, + ppid); +} + +/**************************************************************************** + Actual function that gets POSIX locks. Copes with 64 -> 32 bit cruft and + broken NFS implementations. +****************************************************************************/ + +static bool posix_fcntl_getlock(files_struct *fsp, off_t *poffset, off_t *pcount, int *ptype) +{ + pid_t pid; + bool ret; + + DEBUG(8, ("posix_fcntl_getlock %d %ju %ju %d\n", + fsp_get_io_fd(fsp), (uintmax_t)*poffset, (uintmax_t)*pcount, + *ptype)); + + ret = SMB_VFS_GETLOCK(fsp, poffset, pcount, ptype, &pid); + + if (!ret && ((errno == EFBIG) || (errno == ENOLCK) || (errno == EINVAL))) { + + DEBUG(0, ("posix_fcntl_getlock: WARNING: lock request at " + "offset %ju, length %ju returned\n", + (uintmax_t)*poffset, (uintmax_t)*pcount)); + DEBUGADD(0, ("an %s error. This can happen when using 64 bit " + "lock offsets\n", strerror(errno))); + DEBUGADD(0, ("on 32 bit NFS mounted file systems.\n")); + + /* + * If the offset is > 0x7FFFFFFF then this will cause problems on + * 32 bit NFS mounted filesystems. Just ignore it. + */ + + if (*poffset & ~((off_t)0x7fffffff)) { + DEBUG(0,("Offset greater than 31 bits. Returning success.\n")); + return True; + } + + if (*pcount & ~((off_t)0x7fffffff)) { + /* 32 bit NFS file system, retry with smaller offset */ + DEBUG(0,("Count greater than 31 bits - retrying with 31 bit truncated length.\n")); + errno = 0; + *pcount &= 0x7fffffff; + ret = SMB_VFS_GETLOCK(fsp,poffset,pcount,ptype,&pid); + } + } + + DEBUG(8,("posix_fcntl_getlock: Lock query call %s\n", ret ? "successful" : "failed")); + return ret; +} + +/**************************************************************************** + POSIX function to see if a file region is locked. Returns True if the + region is locked, False otherwise. +****************************************************************************/ + +bool is_posix_locked(files_struct *fsp, + uint64_t *pu_offset, + uint64_t *pu_count, + enum brl_type *plock_type, + enum brl_flavour lock_flav) +{ + off_t offset; + off_t count; + int posix_lock_type = map_posix_lock_type(fsp,*plock_type); + + DEBUG(10, ("is_posix_locked: File %s, offset = %ju, count = %ju, " + "type = %s\n", fsp_str_dbg(fsp), (uintmax_t)*pu_offset, + (uintmax_t)*pu_count, posix_lock_type_name(*plock_type))); + + /* + * If the requested lock won't fit in the POSIX range, we will + * never set it, so presume it is not locked. + */ + + if(!posix_lock_in_range(&offset, &count, *pu_offset, *pu_count)) { + return False; + } + + if (!posix_fcntl_getlock(fsp,&offset,&count,&posix_lock_type)) { + return False; + } + + if (posix_lock_type == F_UNLCK) { + return False; + } + + if (lock_flav == POSIX_LOCK) { + /* Only POSIX lock queries need to know the details. */ + *pu_offset = (uint64_t)offset; + *pu_count = (uint64_t)count; + *plock_type = (posix_lock_type == F_RDLCK) ? READ_LOCK : WRITE_LOCK; + } + return True; +} + +/**************************************************************************** + Next - the functions that deal with in memory database storing representations + of either Windows CIFS locks or POSIX CIFS locks. +****************************************************************************/ + +/* The key used in the in-memory POSIX databases. */ + +struct lock_ref_count_key { + struct file_id id; + char r; +}; + +/******************************************************************* + Form a static locking key for a dev/inode pair for the lock ref count +******************************************************************/ + +static TDB_DATA locking_ref_count_key_fsp(const files_struct *fsp, + struct lock_ref_count_key *tmp) +{ + ZERO_STRUCTP(tmp); + tmp->id = fsp->file_id; + tmp->r = 'r'; + return make_tdb_data((uint8_t *)tmp, sizeof(*tmp)); +} + +/******************************************************************* + Convenience function to get an fd_array key from an fsp. +******************************************************************/ + +static TDB_DATA fd_array_key_fsp(const files_struct *fsp) +{ + return make_tdb_data((const uint8_t *)&fsp->file_id, sizeof(fsp->file_id)); +} + +/******************************************************************* + Create the in-memory POSIX lock databases. +********************************************************************/ + +bool posix_locking_init(bool read_only) +{ + if (posix_pending_close_db != NULL) { + return true; + } + + posix_pending_close_db = db_open_rbt(NULL); + + if (posix_pending_close_db == NULL) { + DEBUG(0,("Failed to open POSIX pending close database.\n")); + return false; + } + + return true; +} + +/******************************************************************* + Delete the in-memory POSIX lock databases. +********************************************************************/ + +bool posix_locking_end(void) +{ + /* + * Shouldn't we close all fd's here? + */ + TALLOC_FREE(posix_pending_close_db); + return true; +} + +/**************************************************************************** + Next - the functions that deal with reference count of number of locks open + on a dev/ino pair. +****************************************************************************/ + +/**************************************************************************** + Increase the lock ref count. Creates lock_ref_count entry if it doesn't exist. +****************************************************************************/ + +static void increment_lock_ref_count(const files_struct *fsp) +{ + struct lock_ref_count_key tmp; + int32_t lock_ref_count = 0; + NTSTATUS status; + + status = dbwrap_change_int32_atomic( + posix_pending_close_db, locking_ref_count_key_fsp(fsp, &tmp), + &lock_ref_count, 1); + + SMB_ASSERT(NT_STATUS_IS_OK(status)); + SMB_ASSERT(lock_ref_count < INT32_MAX); + + DEBUG(10,("lock_ref_count for file %s = %d\n", + fsp_str_dbg(fsp), (int)(lock_ref_count + 1))); +} + +/**************************************************************************** + Reduce the lock ref count. +****************************************************************************/ + +static void decrement_lock_ref_count(const files_struct *fsp) +{ + struct lock_ref_count_key tmp; + int32_t lock_ref_count = 0; + NTSTATUS status; + + status = dbwrap_change_int32_atomic( + posix_pending_close_db, locking_ref_count_key_fsp(fsp, &tmp), + &lock_ref_count, -1); + + SMB_ASSERT(NT_STATUS_IS_OK(status)); + SMB_ASSERT(lock_ref_count > 0); + + DEBUG(10,("lock_ref_count for file %s = %d\n", + fsp_str_dbg(fsp), (int)(lock_ref_count - 1))); +} + +/**************************************************************************** + Fetch the lock ref count. +****************************************************************************/ + +static int32_t get_lock_ref_count(const files_struct *fsp) +{ + struct lock_ref_count_key tmp; + NTSTATUS status; + int32_t lock_ref_count = 0; + + status = dbwrap_fetch_int32( + posix_pending_close_db, locking_ref_count_key_fsp(fsp, &tmp), + &lock_ref_count); + + if (!NT_STATUS_IS_OK(status) && + !NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND)) { + DEBUG(0, ("Error fetching " + "lock ref count for file %s: %s\n", + fsp_str_dbg(fsp), nt_errstr(status))); + } + return lock_ref_count; +} + +/**************************************************************************** + Delete a lock_ref_count entry. +****************************************************************************/ + +static void delete_lock_ref_count(const files_struct *fsp) +{ + struct lock_ref_count_key tmp; + + /* Not a bug if it doesn't exist - no locks were ever granted. */ + + dbwrap_delete(posix_pending_close_db, + locking_ref_count_key_fsp(fsp, &tmp)); + + DEBUG(10,("delete_lock_ref_count for file %s\n", + fsp_str_dbg(fsp))); +} + +/**************************************************************************** + Next - the functions that deal with storing fd's that have outstanding + POSIX locks when closed. +****************************************************************************/ + +/**************************************************************************** + The records in posix_pending_close_db are composed of an array of + ints keyed by dev/ino pair. Those ints are the fd's that were open on + this dev/ino pair that should have been closed, but can't as the lock + ref count is non zero. +****************************************************************************/ + +struct add_fd_to_close_entry_state { + const struct files_struct *fsp; +}; + +static void add_fd_to_close_entry_fn( + struct db_record *rec, + TDB_DATA value, + void *private_data) +{ + struct add_fd_to_close_entry_state *state = private_data; + int fd = fsp_get_pathref_fd(state->fsp); + TDB_DATA values[] = { + value, + { .dptr = (uint8_t *)&fd, + .dsize = sizeof(fd) }, + }; + NTSTATUS status; + + SMB_ASSERT((values[0].dsize % sizeof(int)) == 0); + + status = dbwrap_record_storev(rec, values, ARRAY_SIZE(values), 0); + SMB_ASSERT(NT_STATUS_IS_OK(status)); +} + +/**************************************************************************** + Add an fd to the pending close db. +****************************************************************************/ + +static void add_fd_to_close_entry(const files_struct *fsp) +{ + struct add_fd_to_close_entry_state state = { .fsp = fsp }; + NTSTATUS status; + + status = dbwrap_do_locked( + posix_pending_close_db, + fd_array_key_fsp(fsp), + add_fd_to_close_entry_fn, + &state); + SMB_ASSERT(NT_STATUS_IS_OK(status)); + + DBG_DEBUG("added fd %d file %s\n", + fsp_get_pathref_fd(fsp), + fsp_str_dbg(fsp)); +} + +static void fd_close_posix_fn( + struct db_record *rec, + TDB_DATA data, + void *private_data) +{ + size_t num_fds, i; + + SMB_ASSERT((data.dsize % sizeof(int)) == 0); + num_fds = data.dsize / sizeof(int); + + for (i=0; i<num_fds; i++) { + int fd; + memcpy(&fd, data.dptr, sizeof(int)); + close(fd); + data.dptr += sizeof(int); + } + dbwrap_record_delete(rec); +} + +/**************************************************************************** + Deal with pending closes needed by POSIX locking support. + Note that locking_close_file() is expected to have been called + to delete all locks on this fsp before this function is called. +****************************************************************************/ + +int fd_close_posix(const struct files_struct *fsp) +{ + NTSTATUS status; + + if (!lp_locking(fsp->conn->params) || + !lp_posix_locking(fsp->conn->params) || + fsp->fsp_flags.use_ofd_locks) + { + /* + * No locking or POSIX to worry about or we are using POSIX + * open file description lock semantics which only removes + * locks on the file descriptor we're closing. Just close. + */ + return close(fsp_get_pathref_fd(fsp)); + } + + if (get_lock_ref_count(fsp)) { + + /* + * There are outstanding locks on this dev/inode pair on + * other fds. Add our fd to the pending close db. We also + * set fsp_get_io_fd(fsp) to -1 inside fd_close() after returning + * from VFS layer. + */ + + add_fd_to_close_entry(fsp); + return 0; + } + + status = dbwrap_do_locked( + posix_pending_close_db, + fd_array_key_fsp(fsp), + fd_close_posix_fn, + NULL); + if (!NT_STATUS_IS_OK(status)) { + DBG_WARNING("dbwrap_do_locked failed: %s\n", + nt_errstr(status)); + } + + /* Don't need a lock ref count on this dev/ino anymore. */ + delete_lock_ref_count(fsp); + + /* + * Finally close the fd associated with this fsp. + */ + + return close(fsp_get_pathref_fd(fsp)); +} + +/**************************************************************************** + Next - the functions that deal with the mapping CIFS Windows locks onto + the underlying system POSIX locks. +****************************************************************************/ + +/* + * Structure used when splitting a lock range + * into a POSIX lock range. Doubly linked list. + */ + +struct lock_list { + struct lock_list *next; + struct lock_list *prev; + off_t start; + off_t size; +}; + +/**************************************************************************** + Create a list of lock ranges that don't overlap a given range. Used in calculating + POSIX locks and unlocks. This is a difficult function that requires ASCII art to + understand it :-). +****************************************************************************/ + +static struct lock_list *posix_lock_list(TALLOC_CTX *ctx, + struct lock_list *lhead, + const struct lock_context *lock_ctx, /* Lock context lhead belongs to. */ + const struct lock_struct *plocks, + int num_locks) +{ + int i; + + /* + * Check the current lock list on this dev/inode pair. + * Quit if the list is deleted. + */ + + DEBUG(10, ("posix_lock_list: curr: start=%ju,size=%ju\n", + (uintmax_t)lhead->start, (uintmax_t)lhead->size )); + + for (i=0; i<num_locks && lhead; i++) { + const struct lock_struct *lock = &plocks[i]; + struct lock_list *l_curr; + + /* Ignore all but read/write locks. */ + if (lock->lock_type != READ_LOCK && lock->lock_type != WRITE_LOCK) { + continue; + } + + /* Ignore locks not owned by this process. */ + if (!server_id_equal(&lock->context.pid, &lock_ctx->pid)) { + continue; + } + + /* + * Walk the lock list, checking for overlaps. Note that + * the lock list can expand within this loop if the current + * range being examined needs to be split. + */ + + for (l_curr = lhead; l_curr;) { + + DEBUG(10, ("posix_lock_list: lock: fnum=%ju: " + "start=%ju,size=%ju:type=%s", + (uintmax_t)lock->fnum, + (uintmax_t)lock->start, + (uintmax_t)lock->size, + posix_lock_type_name(lock->lock_type) )); + + if ( (l_curr->start >= (lock->start + lock->size)) || + (lock->start >= (l_curr->start + l_curr->size))) { + + /* No overlap with existing lock - leave this range alone. */ +/********************************************* + +---------+ + | l_curr | + +---------+ + +-------+ + | lock | + +-------+ +OR.... + +---------+ + | l_curr | + +---------+ +**********************************************/ + + DEBUG(10,(" no overlap case.\n" )); + + l_curr = l_curr->next; + + } else if ( (l_curr->start >= lock->start) && + (l_curr->start + l_curr->size <= lock->start + lock->size) ) { + + /* + * This range is completely overlapped by this existing lock range + * and thus should have no effect. Delete it from the list. + */ +/********************************************* + +---------+ + | l_curr | + +---------+ + +---------------------------+ + | lock | + +---------------------------+ +**********************************************/ + /* Save the next pointer */ + struct lock_list *ul_next = l_curr->next; + + DEBUG(10,(" delete case.\n" )); + + DLIST_REMOVE(lhead, l_curr); + if(lhead == NULL) { + break; /* No more list... */ + } + + l_curr = ul_next; + + } else if ( (l_curr->start >= lock->start) && + (l_curr->start < lock->start + lock->size) && + (l_curr->start + l_curr->size > lock->start + lock->size) ) { + + /* + * This range overlaps the existing lock range at the high end. + * Truncate by moving start to existing range end and reducing size. + */ +/********************************************* + +---------------+ + | l_curr | + +---------------+ + +---------------+ + | lock | + +---------------+ +BECOMES.... + +-------+ + | l_curr| + +-------+ +**********************************************/ + + l_curr->size = (l_curr->start + l_curr->size) - (lock->start + lock->size); + l_curr->start = lock->start + lock->size; + + DEBUG(10, (" truncate high case: start=%ju," + "size=%ju\n", + (uintmax_t)l_curr->start, + (uintmax_t)l_curr->size )); + + l_curr = l_curr->next; + + } else if ( (l_curr->start < lock->start) && + (l_curr->start + l_curr->size > lock->start) && + (l_curr->start + l_curr->size <= lock->start + lock->size) ) { + + /* + * This range overlaps the existing lock range at the low end. + * Truncate by reducing size. + */ +/********************************************* + +---------------+ + | l_curr | + +---------------+ + +---------------+ + | lock | + +---------------+ +BECOMES.... + +-------+ + | l_curr| + +-------+ +**********************************************/ + + l_curr->size = lock->start - l_curr->start; + + DEBUG(10, (" truncate low case: start=%ju," + "size=%ju\n", + (uintmax_t)l_curr->start, + (uintmax_t)l_curr->size )); + + l_curr = l_curr->next; + + } else if ( (l_curr->start < lock->start) && + (l_curr->start + l_curr->size > lock->start + lock->size) ) { + /* + * Worst case scenario. Range completely overlaps an existing + * lock range. Split the request into two, push the new (upper) request + * into the dlink list, and continue with the entry after l_new (as we + * know that l_new will not overlap with this lock). + */ +/********************************************* + +---------------------------+ + | l_curr | + +---------------------------+ + +---------+ + | lock | + +---------+ +BECOMES..... + +-------+ +---------+ + | l_curr| | l_new | + +-------+ +---------+ +**********************************************/ + struct lock_list *l_new = talloc(ctx, struct lock_list); + + if(l_new == NULL) { + DEBUG(0,("posix_lock_list: talloc fail.\n")); + return NULL; /* The talloc_destroy takes care of cleanup. */ + } + + ZERO_STRUCTP(l_new); + l_new->start = lock->start + lock->size; + l_new->size = l_curr->start + l_curr->size - l_new->start; + + /* Truncate the l_curr. */ + l_curr->size = lock->start - l_curr->start; + + DEBUG(10, (" split case: curr: start=%ju," + "size=%ju new: start=%ju," + "size=%ju\n", + (uintmax_t)l_curr->start, + (uintmax_t)l_curr->size, + (uintmax_t)l_new->start, + (uintmax_t)l_new->size )); + + /* + * Add into the dlink list after the l_curr point - NOT at lhead. + */ + DLIST_ADD_AFTER(lhead, l_new, l_curr); + + /* And move after the link we added. */ + l_curr = l_new->next; + + } else { + + /* + * This logic case should never happen. Ensure this is the + * case by forcing an abort.... Remove in production. + */ + char *msg = NULL; + + if (asprintf(&msg, "logic flaw in cases: " + "l_curr: start = %ju, " + "size = %ju : lock: " + "start = %ju, size = %ju", + (uintmax_t)l_curr->start, + (uintmax_t)l_curr->size, + (uintmax_t)lock->start, + (uintmax_t)lock->size ) != -1) { + smb_panic(msg); + } else { + smb_panic("posix_lock_list"); + } + } + } /* end for ( l_curr = lhead; l_curr;) */ + } /* end for (i=0; i<num_locks && ul_head; i++) */ + + return lhead; +} + +/**************************************************************************** + POSIX function to acquire a lock. Returns True if the + lock could be granted, False if not. +****************************************************************************/ + +bool set_posix_lock_windows_flavour(files_struct *fsp, + uint64_t u_offset, + uint64_t u_count, + enum brl_type lock_type, + const struct lock_context *lock_ctx, + const struct lock_struct *plocks, + int num_locks, + int *errno_ret) +{ + off_t offset; + off_t count; + int posix_lock_type = map_posix_lock_type(fsp,lock_type); + bool ret = True; + size_t lock_count; + TALLOC_CTX *l_ctx = NULL; + struct lock_list *llist = NULL; + struct lock_list *ll = NULL; + + DEBUG(5, ("set_posix_lock_windows_flavour: File %s, offset = %ju, " + "count = %ju, type = %s\n", fsp_str_dbg(fsp), + (uintmax_t)u_offset, (uintmax_t)u_count, + posix_lock_type_name(lock_type))); + + /* + * If the requested lock won't fit in the POSIX range, we will + * pretend it was successful. + */ + + if(!posix_lock_in_range(&offset, &count, u_offset, u_count)) { + increment_lock_ref_count(fsp); + return True; + } + + /* + * Windows is very strange. It allows read locks to be overlayed + * (even over a write lock), but leaves the write lock in force until the first + * unlock. It also reference counts the locks. This means the following sequence : + * + * process1 process2 + * ------------------------------------------------------------------------ + * WRITE LOCK : start = 2, len = 10 + * READ LOCK: start =0, len = 10 - FAIL + * READ LOCK : start = 0, len = 14 + * READ LOCK: start =0, len = 10 - FAIL + * UNLOCK : start = 2, len = 10 + * READ LOCK: start =0, len = 10 - OK + * + * Under POSIX, the same sequence in steps 1 and 2 would not be reference counted, but + * would leave a single read lock over the 0-14 region. + */ + + if ((l_ctx = talloc_init("set_posix_lock")) == NULL) { + DEBUG(0,("set_posix_lock_windows_flavour: unable to init talloc context.\n")); + return False; + } + + if ((ll = talloc(l_ctx, struct lock_list)) == NULL) { + DEBUG(0,("set_posix_lock_windows_flavour: unable to talloc unlock list.\n")); + talloc_destroy(l_ctx); + return False; + } + + /* + * Create the initial list entry containing the + * lock we want to add. + */ + + ZERO_STRUCTP(ll); + ll->start = offset; + ll->size = count; + + DLIST_ADD(llist, ll); + + /* + * The following call calculates if there are any + * overlapping locks held by this process on + * fd's open on the same file and splits this list + * into a list of lock ranges that do not overlap with existing + * POSIX locks. + */ + + llist = posix_lock_list(l_ctx, + llist, + lock_ctx, /* Lock context llist belongs to. */ + plocks, + num_locks); + + /* + * Add the POSIX locks on the list of ranges returned. + * As the lock is supposed to be added atomically, we need to + * back out all the locks if any one of these calls fail. + */ + + for (lock_count = 0, ll = llist; ll; ll = ll->next, lock_count++) { + offset = ll->start; + count = ll->size; + + DEBUG(5, ("set_posix_lock_windows_flavour: Real lock: " + "Type = %s: offset = %ju, count = %ju\n", + posix_lock_type_name(posix_lock_type), + (uintmax_t)offset, (uintmax_t)count )); + + if (!posix_fcntl_lock(fsp,F_SETLK,offset,count,posix_lock_type)) { + *errno_ret = errno; + DEBUG(5, ("set_posix_lock_windows_flavour: Lock " + "fail !: Type = %s: offset = %ju, " + "count = %ju. Errno = %s\n", + posix_lock_type_name(posix_lock_type), + (uintmax_t)offset, (uintmax_t)count, + strerror(errno) )); + ret = False; + break; + } + } + + if (!ret) { + + /* + * Back out all the POSIX locks we have on fail. + */ + + for (ll = llist; lock_count; ll = ll->next, lock_count--) { + offset = ll->start; + count = ll->size; + + DEBUG(5, ("set_posix_lock_windows_flavour: Backing " + "out locks: Type = %s: offset = %ju, " + "count = %ju\n", + posix_lock_type_name(posix_lock_type), + (uintmax_t)offset, (uintmax_t)count )); + + posix_fcntl_lock(fsp,F_SETLK,offset,count,F_UNLCK); + } + } else { + /* Remember the number of locks we have on this dev/ino pair. */ + increment_lock_ref_count(fsp); + } + + talloc_destroy(l_ctx); + return ret; +} + +/**************************************************************************** + POSIX function to release a lock. Returns True if the + lock could be released, False if not. +****************************************************************************/ + +bool release_posix_lock_windows_flavour(files_struct *fsp, + uint64_t u_offset, + uint64_t u_count, + enum brl_type deleted_lock_type, + const struct lock_context *lock_ctx, + const struct lock_struct *plocks, + int num_locks) +{ + off_t offset; + off_t count; + bool ret = True; + TALLOC_CTX *ul_ctx = NULL; + struct lock_list *ulist = NULL; + struct lock_list *ul = NULL; + + DEBUG(5, ("release_posix_lock_windows_flavour: File %s, offset = %ju, " + "count = %ju\n", fsp_str_dbg(fsp), + (uintmax_t)u_offset, (uintmax_t)u_count)); + + /* Remember the number of locks we have on this dev/ino pair. */ + decrement_lock_ref_count(fsp); + + /* + * If the requested lock won't fit in the POSIX range, we will + * pretend it was successful. + */ + + if(!posix_lock_in_range(&offset, &count, u_offset, u_count)) { + return True; + } + + if ((ul_ctx = talloc_init("release_posix_lock")) == NULL) { + DEBUG(0,("release_posix_lock_windows_flavour: unable to init talloc context.\n")); + return False; + } + + if ((ul = talloc(ul_ctx, struct lock_list)) == NULL) { + DEBUG(0,("release_posix_lock_windows_flavour: unable to talloc unlock list.\n")); + talloc_destroy(ul_ctx); + return False; + } + + /* + * Create the initial list entry containing the + * lock we want to remove. + */ + + ZERO_STRUCTP(ul); + ul->start = offset; + ul->size = count; + + DLIST_ADD(ulist, ul); + + /* + * The following call calculates if there are any + * overlapping locks held by this process on + * fd's open on the same file and creates a + * list of unlock ranges that will allow + * POSIX lock ranges to remain on the file whilst the + * unlocks are performed. + */ + + ulist = posix_lock_list(ul_ctx, + ulist, + lock_ctx, /* Lock context ulist belongs to. */ + plocks, + num_locks); + + /* + * If there were any overlapped entries (list is > 1 or size or start have changed), + * and the lock_type we just deleted from + * the upper layer tdb was a write lock, then before doing the unlock we need to downgrade + * the POSIX lock to a read lock. This allows any overlapping read locks + * to be atomically maintained. + */ + + if (deleted_lock_type == WRITE_LOCK && + (!ulist || ulist->next != NULL || ulist->start != offset || ulist->size != count)) { + + DEBUG(5, ("release_posix_lock_windows_flavour: downgrading " + "lock to READ: offset = %ju, count = %ju\n", + (uintmax_t)offset, (uintmax_t)count )); + + if (!posix_fcntl_lock(fsp,F_SETLK,offset,count,F_RDLCK)) { + DEBUG(0,("release_posix_lock_windows_flavour: downgrade of lock failed with error %s !\n", strerror(errno) )); + talloc_destroy(ul_ctx); + return False; + } + } + + /* + * Release the POSIX locks on the list of ranges returned. + */ + + for(; ulist; ulist = ulist->next) { + offset = ulist->start; + count = ulist->size; + + DEBUG(5, ("release_posix_lock_windows_flavour: Real unlock: " + "offset = %ju, count = %ju\n", + (uintmax_t)offset, (uintmax_t)count )); + + if (!posix_fcntl_lock(fsp,F_SETLK,offset,count,F_UNLCK)) { + ret = False; + } + } + + talloc_destroy(ul_ctx); + return ret; +} + +/**************************************************************************** + Next - the functions that deal with mapping CIFS POSIX locks onto + the underlying system POSIX locks. +****************************************************************************/ + +/**************************************************************************** + We only increment the lock ref count when we see a POSIX lock on a context + that doesn't already have them. +****************************************************************************/ + +static void increment_posix_lock_count(const files_struct *fsp, + uint64_t smblctx) +{ + NTSTATUS status; + TDB_DATA ctx_key; + TDB_DATA val = { 0 }; + + ctx_key.dptr = (uint8_t *)&smblctx; + ctx_key.dsize = sizeof(smblctx); + + /* + * Don't increment if we already have any POSIX flavor + * locks on this context. + */ + if (dbwrap_exists(posix_pending_close_db, ctx_key)) { + return; + } + + /* Remember that we have POSIX flavor locks on this context. */ + status = dbwrap_store(posix_pending_close_db, ctx_key, val, 0); + SMB_ASSERT(NT_STATUS_IS_OK(status)); + + increment_lock_ref_count(fsp); + + DEBUG(10,("posix_locks set for file %s\n", + fsp_str_dbg(fsp))); +} + +static void decrement_posix_lock_count(const files_struct *fsp, uint64_t smblctx) +{ + NTSTATUS status; + TDB_DATA ctx_key; + + ctx_key.dptr = (uint8_t *)&smblctx; + ctx_key.dsize = sizeof(smblctx); + + status = dbwrap_delete(posix_pending_close_db, ctx_key); + SMB_ASSERT(NT_STATUS_IS_OK(status)); + + decrement_lock_ref_count(fsp); + + DEBUG(10,("posix_locks deleted for file %s\n", + fsp_str_dbg(fsp))); +} + +/**************************************************************************** + Return true if any locks exist on the given lock context. +****************************************************************************/ + +static bool locks_exist_on_context(const struct lock_struct *plocks, + int num_locks, + const struct lock_context *lock_ctx) +{ + int i; + + for (i=0; i < num_locks; i++) { + const struct lock_struct *lock = &plocks[i]; + + /* Ignore all but read/write locks. */ + if (lock->lock_type != READ_LOCK && lock->lock_type != WRITE_LOCK) { + continue; + } + + /* Ignore locks not owned by this process. */ + if (!server_id_equal(&lock->context.pid, &lock_ctx->pid)) { + continue; + } + + if (lock_ctx->smblctx == lock->context.smblctx) { + return true; + } + } + return false; +} + +/**************************************************************************** + POSIX function to acquire a lock. Returns True if the + lock could be granted, False if not. + As POSIX locks don't stack or conflict (they just overwrite) + we can map the requested lock directly onto a system one. We + know it doesn't conflict with locks on other contexts as the + upper layer would have refused it. +****************************************************************************/ + +bool set_posix_lock_posix_flavour(files_struct *fsp, + uint64_t u_offset, + uint64_t u_count, + enum brl_type lock_type, + const struct lock_context *lock_ctx, + int *errno_ret) +{ + off_t offset; + off_t count; + int posix_lock_type = map_posix_lock_type(fsp,lock_type); + + DEBUG(5,("set_posix_lock_posix_flavour: File %s, offset = %ju, count " + "= %ju, type = %s\n", fsp_str_dbg(fsp), + (uintmax_t)u_offset, (uintmax_t)u_count, + posix_lock_type_name(lock_type))); + + /* + * If the requested lock won't fit in the POSIX range, we will + * pretend it was successful. + */ + + if(!posix_lock_in_range(&offset, &count, u_offset, u_count)) { + increment_posix_lock_count(fsp, lock_ctx->smblctx); + return True; + } + + if (!posix_fcntl_lock(fsp,F_SETLK,offset,count,posix_lock_type)) { + *errno_ret = errno; + DEBUG(5,("set_posix_lock_posix_flavour: Lock fail !: Type = %s: offset = %ju, count = %ju. Errno = %s\n", + posix_lock_type_name(posix_lock_type), (intmax_t)offset, (intmax_t)count, strerror(errno) )); + return False; + } + increment_posix_lock_count(fsp, lock_ctx->smblctx); + return True; +} + +/**************************************************************************** + POSIX function to release a lock. Returns True if the + lock could be released, False if not. + We are given a complete lock state from the upper layer which is what the lock + state should be after the unlock has already been done, so what + we do is punch out holes in the unlock range where locks owned by this process + have a different lock context. +****************************************************************************/ + +bool release_posix_lock_posix_flavour(files_struct *fsp, + uint64_t u_offset, + uint64_t u_count, + const struct lock_context *lock_ctx, + const struct lock_struct *plocks, + int num_locks) +{ + bool ret = True; + off_t offset; + off_t count; + TALLOC_CTX *ul_ctx = NULL; + struct lock_list *ulist = NULL; + struct lock_list *ul = NULL; + + DEBUG(5, ("release_posix_lock_posix_flavour: File %s, offset = %ju, " + "count = %ju\n", fsp_str_dbg(fsp), + (uintmax_t)u_offset, (uintmax_t)u_count)); + + /* + * If the requested lock won't fit in the POSIX range, we will + * pretend it was successful. + */ + + if(!posix_lock_in_range(&offset, &count, u_offset, u_count)) { + if (!locks_exist_on_context(plocks, num_locks, lock_ctx)) { + decrement_posix_lock_count(fsp, lock_ctx->smblctx); + } + return True; + } + + if ((ul_ctx = talloc_init("release_posix_lock")) == NULL) { + DEBUG(0,("release_posix_lock_windows_flavour: unable to init talloc context.\n")); + return False; + } + + if ((ul = talloc(ul_ctx, struct lock_list)) == NULL) { + DEBUG(0,("release_posix_lock_windows_flavour: unable to talloc unlock list.\n")); + talloc_destroy(ul_ctx); + return False; + } + + /* + * Create the initial list entry containing the + * lock we want to remove. + */ + + ZERO_STRUCTP(ul); + ul->start = offset; + ul->size = count; + + DLIST_ADD(ulist, ul); + + /* + * Walk the given array creating a linked list + * of unlock requests. + */ + + ulist = posix_lock_list(ul_ctx, + ulist, + lock_ctx, /* Lock context ulist belongs to. */ + plocks, + num_locks); + + /* + * Release the POSIX locks on the list of ranges returned. + */ + + for(; ulist; ulist = ulist->next) { + offset = ulist->start; + count = ulist->size; + + DEBUG(5, ("release_posix_lock_posix_flavour: Real unlock: " + "offset = %ju, count = %ju\n", + (uintmax_t)offset, (uintmax_t)count )); + + if (!posix_fcntl_lock(fsp,F_SETLK,offset,count,F_UNLCK)) { + ret = False; + } + } + + if (!locks_exist_on_context(plocks, num_locks, lock_ctx)) { + decrement_posix_lock_count(fsp, lock_ctx->smblctx); + } + talloc_destroy(ul_ctx); + return ret; +} diff --git a/source3/locking/proto.h b/source3/locking/proto.h new file mode 100644 index 0000000..7fc177d --- /dev/null +++ b/source3/locking/proto.h @@ -0,0 +1,215 @@ +/* + * Unix SMB/CIFS implementation. + * Locking functions + * + * Copyright (C) Andrew Tridgell 1992-2000 + * Copyright (C) Jeremy Allison 1992-2006 + * Copyright (C) Volker Lendecke 2005 + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 3 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program. If not, see <http://www.gnu.org/licenses/>. + */ + +#ifndef _LOCKING_PROTO_H_ +#define _LOCKING_PROTO_H_ + +#include <tdb.h> + +/* The following definitions come from locking/brlock.c */ + +void brl_init(bool read_only); +void brl_shutdown(void); + +unsigned int brl_num_locks(const struct byte_range_lock *brl); +struct files_struct *brl_fsp(struct byte_range_lock *brl); +TALLOC_CTX *brl_req_mem_ctx(const struct byte_range_lock *brl); +const struct GUID *brl_req_guid(const struct byte_range_lock *brl); + +bool byte_range_valid(uint64_t ofs, uint64_t len); +bool byte_range_overlap(uint64_t ofs1, + uint64_t len1, + uint64_t ofs2, + uint64_t len2); + +NTSTATUS brl_lock_windows_default(struct byte_range_lock *br_lck, + struct lock_struct *plock); + +NTSTATUS brl_lock( + struct byte_range_lock *br_lck, + uint64_t smblctx, + struct server_id pid, + br_off start, + br_off size, + enum brl_type lock_type, + enum brl_flavour lock_flav, + struct server_id *blocker_pid, + uint64_t *psmblctx); +bool brl_unlock(struct byte_range_lock *br_lck, + uint64_t smblctx, + struct server_id pid, + br_off start, + br_off size, + enum brl_flavour lock_flav); +bool brl_unlock_windows_default(struct byte_range_lock *br_lck, + const struct lock_struct *plock); +bool brl_locktest(struct byte_range_lock *br_lck, + const struct lock_struct *rw_probe); +NTSTATUS brl_lockquery(struct byte_range_lock *br_lck, + uint64_t *psmblctx, + struct server_id pid, + br_off *pstart, + br_off *psize, + enum brl_type *plock_type, + enum brl_flavour lock_flav); +bool brl_mark_disconnected(struct files_struct *fsp); +bool brl_reconnect_disconnected(struct files_struct *fsp); +void brl_close_fnum(struct byte_range_lock *br_lck); +int brl_forall(void (*fn)(struct file_id id, struct server_id pid, + enum brl_type lock_type, + enum brl_flavour lock_flav, + br_off start, br_off size, + void *private_data), + void *private_data); +struct byte_range_lock *brl_get_locks_for_locking(TALLOC_CTX *mem_ctx, + files_struct *fsp, + TALLOC_CTX *req_mem_ctx, + const struct GUID *req_guid); +struct byte_range_lock *brl_get_locks(TALLOC_CTX *mem_ctx, + files_struct *fsp); +struct byte_range_lock *brl_get_locks_readonly(files_struct *fsp); +bool brl_cleanup_disconnected(struct file_id fid, uint64_t open_persistent_id); + +/* The following definitions come from locking/locking.c */ + +const char *lock_type_name(enum brl_type lock_type); +const char *lock_flav_name(enum brl_flavour lock_flav); +void init_strict_lock_struct(files_struct *fsp, + uint64_t smblctx, + br_off start, + br_off size, + enum brl_type lock_type, + enum brl_flavour lock_flav, + struct lock_struct *plock); +bool strict_lock_check_default(files_struct *fsp, + struct lock_struct *plock); +NTSTATUS query_lock(files_struct *fsp, + uint64_t *psmblctx, + uint64_t *pcount, + uint64_t *poffset, + enum brl_type *plock_type, + enum brl_flavour lock_flav); +NTSTATUS do_lock(files_struct *fsp, + TALLOC_CTX *req_mem_ctx, + const struct GUID *req_guid, + uint64_t smblctx, + uint64_t count, + uint64_t offset, + enum brl_type lock_type, + enum brl_flavour lock_flav, + struct server_id *pblocker_pid, + uint64_t *psmblctx); +NTSTATUS do_unlock(files_struct *fsp, + uint64_t smblctx, + uint64_t count, + uint64_t offset, + enum brl_flavour lock_flav); +void locking_close_file(files_struct *fsp, + enum file_close_type close_type); +char *share_mode_str(TALLOC_CTX *ctx, int num, + const struct file_id *id, + const struct share_mode_entry *e); + +bool rename_share_filename(struct messaging_context *msg_ctx, + struct share_mode_lock *lck, + struct file_id id, + const char *servicepath, + uint32_t orig_name_hash, + uint32_t new_name_hash, + const struct smb_filename *smb_fname); +void get_file_infos(struct file_id id, + uint32_t name_hash, + bool *delete_on_close, + struct timespec *write_time); +bool is_valid_share_mode_entry(const struct share_mode_entry *e); +bool share_entry_stale_pid(struct share_mode_entry *e); +NTSTATUS remove_lease_if_stale(struct share_mode_lock *lck, + const struct GUID *client_guid, + const struct smb2_lease_key *lease_key); +bool get_delete_on_close_token(struct share_mode_lock *lck, + uint32_t name_hash, + const struct security_token **pp_nt_tok, + const struct security_unix_token **pp_tok); +void reset_delete_on_close_lck(files_struct *fsp, + struct share_mode_lock *lck); +void set_delete_on_close_lck(files_struct *fsp, + struct share_mode_lock *lck, + const struct security_token *nt_tok, + const struct security_unix_token *tok); +bool set_delete_on_close(files_struct *fsp, bool delete_on_close, + const struct security_token *nt_tok, + const struct security_unix_token *tok); +bool is_delete_on_close_set(struct share_mode_lock *lck, uint32_t name_hash); +bool set_sticky_write_time(struct file_id fileid, struct timespec write_time); +bool set_write_time(struct file_id fileid, struct timespec write_time); +struct timespec get_share_mode_write_time(struct share_mode_lock *lck); +bool file_has_open_streams(files_struct *fsp); +bool share_mode_forall_leases( + struct share_mode_lock *lck, + bool (*fn)(struct share_mode_entry *e, + void *private_data), + void *private_data); + +/* The following definitions come from locking/posix.c */ + +bool is_posix_locked(files_struct *fsp, + uint64_t *pu_offset, + uint64_t *pu_count, + enum brl_type *plock_type, + enum brl_flavour lock_flav); +bool posix_locking_init(bool read_only); +bool posix_locking_end(void); +int fd_close_posix(const struct files_struct *fsp); +bool set_posix_lock_windows_flavour(files_struct *fsp, + uint64_t u_offset, + uint64_t u_count, + enum brl_type lock_type, + const struct lock_context *lock_ctx, + const struct lock_struct *plocks, + int num_locks, + int *errno_ret); +bool release_posix_lock_windows_flavour(files_struct *fsp, + uint64_t u_offset, + uint64_t u_count, + enum brl_type deleted_lock_type, + const struct lock_context *lock_ctx, + const struct lock_struct *plocks, + int num_locks); +bool set_posix_lock_posix_flavour(files_struct *fsp, + uint64_t u_offset, + uint64_t u_count, + enum brl_type lock_type, + const struct lock_context *lock_ctx, + int *errno_ret); +bool release_posix_lock_posix_flavour(files_struct *fsp, + uint64_t u_offset, + uint64_t u_count, + const struct lock_context *lock_ctx, + const struct lock_struct *plocks, + int num_locks); + +/* The following definitions come from locking/leases_util.c */ +uint32_t map_oplock_to_lease_type(uint16_t op_type); +uint32_t fsp_lease_type(struct files_struct *fsp); +const struct GUID *fsp_client_guid(const files_struct *fsp); + +#endif /* _LOCKING_PROTO_H_ */ diff --git a/source3/locking/share_mode_lock.c b/source3/locking/share_mode_lock.c new file mode 100644 index 0000000..75912ec --- /dev/null +++ b/source3/locking/share_mode_lock.c @@ -0,0 +1,2517 @@ +/* + Unix SMB/CIFS implementation. + Locking functions + Copyright (C) Andrew Tridgell 1992-2000 + Copyright (C) Jeremy Allison 1992-2006 + Copyright (C) Volker Lendecke 2005 + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. + + Revision History: + + 12 aug 96: Erik.Devriendt@te6.siemens.be + added support for shared memory implementation of share mode locking + + May 1997. Jeremy Allison (jallison@whistle.com). Modified share mode + locking to deal with multiple share modes per open file. + + September 1997. Jeremy Allison (jallison@whistle.com). Added oplock + support. + + rewritten completely to use new tdb code. Tridge, Dec '99 + + Added POSIX locking support. Jeremy Allison (jeremy@valinux.com), Apr. 2000. + Added Unix Extensions POSIX locking support. Jeremy Allison Mar 2006. +*/ + +#include "includes.h" +#include "system/filesys.h" +#include "lib/util/server_id.h" +#include "share_mode_lock.h" +#include "share_mode_lock_private.h" +#include "locking/proto.h" +#include "smbd/globals.h" +#include "dbwrap/dbwrap.h" +#include "dbwrap/dbwrap_open.h" +#include "dbwrap/dbwrap_private.h" +#include "../libcli/security/security.h" +#include "serverid.h" +#include "messages.h" +#include "util_tdb.h" +#include "../librpc/gen_ndr/ndr_open_files.h" +#include "source3/lib/dbwrap/dbwrap_watch.h" +#include "locking/leases_db.h" +#include "../lib/util/memcache.h" +#include "lib/util/tevent_ntstatus.h" +#include "g_lock.h" +#include "smbd/fd_handle.h" +#include "lib/global_contexts.h" + +#undef DBGC_CLASS +#define DBGC_CLASS DBGC_LOCKING + +#define NO_LOCKING_COUNT (-1) + +/* the locking database handle */ +static struct g_lock_ctx *lock_ctx; + +static bool locking_init_internal(bool read_only) +{ + struct db_context *backend; + char *db_path; + + brl_init(read_only); + + if (lock_ctx != NULL) { + return True; + } + + db_path = lock_path(talloc_tos(), "locking.tdb"); + if (db_path == NULL) { + return false; + } + + backend = db_open(NULL, db_path, + SMB_OPEN_DATABASE_TDB_HASH_SIZE, + TDB_DEFAULT| + TDB_VOLATILE| + TDB_CLEAR_IF_FIRST| + TDB_INCOMPATIBLE_HASH| + TDB_SEQNUM, + read_only?O_RDONLY:O_RDWR|O_CREAT, 0644, + DBWRAP_LOCK_ORDER_NONE, + DBWRAP_FLAG_NONE); + TALLOC_FREE(db_path); + if (!backend) { + DEBUG(0,("ERROR: Failed to initialise locking database\n")); + return False; + } + + lock_ctx = g_lock_ctx_init_backend( + NULL, global_messaging_context(), &backend); + if (lock_ctx == NULL) { + TALLOC_FREE(backend); + return false; + } + g_lock_set_lock_order(lock_ctx, DBWRAP_LOCK_ORDER_1); + + if (!posix_locking_init(read_only)) { + TALLOC_FREE(lock_ctx); + return False; + } + + return True; +} + +bool locking_init(void) +{ + return locking_init_internal(false); +} + +bool locking_init_readonly(void) +{ + return locking_init_internal(true); +} + +/******************************************************************* + Deinitialize the share_mode management. +******************************************************************/ + +bool locking_end(void) +{ + brl_shutdown(); + TALLOC_FREE(lock_ctx); + return true; +} + +/******************************************************************* + Form a static locking key for a dev/inode pair. +******************************************************************/ + +static TDB_DATA locking_key(const struct file_id *id) +{ + return make_tdb_data((const uint8_t *)id, sizeof(*id)); +} + +/******************************************************************* + Share mode cache utility functions that store/delete/retrieve + entries from memcache. + + For now share the statcache (global cache) memory space. If + a lock record gets orphaned (which shouldn't happen as we're + using the same locking_key data as lookup) it will eventually + fall out of the cache via the normal LRU trim mechanism. If + necessary we can always make this a separate (smaller) cache. +******************************************************************/ + +static DATA_BLOB memcache_key(const struct file_id *id) +{ + return data_blob_const((const void *)id, sizeof(*id)); +} + +static void share_mode_memcache_store(struct share_mode_data *d) +{ + const DATA_BLOB key = memcache_key(&d->id); + struct file_id_buf idbuf; + + DBG_DEBUG("stored entry for file %s epoch %"PRIx64" key %s\n", + d->base_name, + d->unique_content_epoch, + file_id_str_buf(d->id, &idbuf)); + + /* Ensure everything stored in the cache is pristine. */ + d->modified = false; + d->fresh = false; + + /* + * Ensure the memory going into the cache + * doesn't have a destructor so it can be + * cleanly evicted by the memcache LRU + * mechanism. + */ + talloc_set_destructor(d, NULL); + + /* Cache will own d after this call. */ + memcache_add_talloc(NULL, + SHARE_MODE_LOCK_CACHE, + key, + &d); +} + +/* + * NB. We use ndr_pull_hyper on a stack-created + * struct ndr_pull with no talloc allowed, as we + * need this to be really fast as an ndr-peek into + * the first 10 bytes of the blob. + */ + +static enum ndr_err_code get_share_mode_blob_header( + const uint8_t *buf, size_t buflen, uint64_t *pepoch, uint16_t *pflags) +{ + struct ndr_pull ndr = { + .data = discard_const_p(uint8_t, buf), + .data_size = buflen, + }; + NDR_CHECK(ndr_pull_hyper(&ndr, NDR_SCALARS, pepoch)); + NDR_CHECK(ndr_pull_uint16(&ndr, NDR_SCALARS, pflags)); + return NDR_ERR_SUCCESS; +} + +struct fsp_update_share_mode_flags_state { + enum ndr_err_code ndr_err; + uint16_t share_mode_flags; +}; + +static void fsp_update_share_mode_flags_fn( + const uint8_t *buf, + size_t buflen, + bool *modified_dependent, + void *private_data) +{ + struct fsp_update_share_mode_flags_state *state = private_data; + uint64_t seq; + + state->ndr_err = get_share_mode_blob_header( + buf, buflen, &seq, &state->share_mode_flags); +} + +static NTSTATUS fsp_update_share_mode_flags(struct files_struct *fsp) +{ + struct fsp_update_share_mode_flags_state state = {0}; + int seqnum = g_lock_seqnum(lock_ctx); + NTSTATUS status; + + if (seqnum == fsp->share_mode_flags_seqnum) { + return NT_STATUS_OK; + } + + status = share_mode_do_locked( + fsp->file_id, fsp_update_share_mode_flags_fn, &state); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("share_mode_do_locked returned %s\n", + nt_errstr(status)); + return status; + } + + if (!NDR_ERR_CODE_IS_SUCCESS(state.ndr_err)) { + DBG_DEBUG("get_share_mode_blob_header returned %s\n", + ndr_errstr(state.ndr_err)); + return ndr_map_error2ntstatus(state.ndr_err); + } + + fsp->share_mode_flags_seqnum = seqnum; + fsp->share_mode_flags = state.share_mode_flags; + + return NT_STATUS_OK; +} + +bool file_has_read_lease(struct files_struct *fsp) +{ + NTSTATUS status; + + status = fsp_update_share_mode_flags(fsp); + if (!NT_STATUS_IS_OK(status)) { + /* Safe default for leases */ + return true; + } + + return (fsp->share_mode_flags & SHARE_MODE_LEASE_READ) != 0; +} + +static int share_mode_data_nofree_destructor(struct share_mode_data *d) +{ + return -1; +} + +static struct share_mode_data *share_mode_memcache_fetch( + TALLOC_CTX *mem_ctx, + struct file_id id, + const uint8_t *buf, + size_t buflen) +{ + const DATA_BLOB key = memcache_key(&id); + enum ndr_err_code ndr_err; + struct share_mode_data *d; + uint64_t unique_content_epoch; + uint16_t flags; + void *ptr; + struct file_id_buf idbuf; + + ptr = memcache_lookup_talloc(NULL, + SHARE_MODE_LOCK_CACHE, + key); + if (ptr == NULL) { + DBG_DEBUG("failed to find entry for key %s\n", + file_id_str_buf(id, &idbuf)); + return NULL; + } + /* sequence number key is at start of blob. */ + ndr_err = get_share_mode_blob_header( + buf, buflen, &unique_content_epoch, &flags); + if (ndr_err != NDR_ERR_SUCCESS) { + /* Bad blob. Remove entry. */ + DBG_DEBUG("bad blob %u key %s\n", + (unsigned int)ndr_err, + file_id_str_buf(id, &idbuf)); + memcache_delete(NULL, + SHARE_MODE_LOCK_CACHE, + key); + return NULL; + } + + d = (struct share_mode_data *)ptr; + if (d->unique_content_epoch != unique_content_epoch) { + DBG_DEBUG("epoch changed (cached %"PRIx64") (new %"PRIx64") " + "for key %s\n", + d->unique_content_epoch, + unique_content_epoch, + file_id_str_buf(id, &idbuf)); + /* Cache out of date. Remove entry. */ + memcache_delete(NULL, + SHARE_MODE_LOCK_CACHE, + key); + return NULL; + } + + /* Move onto mem_ctx. */ + d = talloc_move(mem_ctx, &ptr); + + /* + * Now we own d, prevent the cache from freeing it + * when we delete the entry. + */ + talloc_set_destructor(d, share_mode_data_nofree_destructor); + + /* Remove from the cache. We own it now. */ + memcache_delete(NULL, + SHARE_MODE_LOCK_CACHE, + key); + + /* And reset the destructor to none. */ + talloc_set_destructor(d, NULL); + + DBG_DEBUG("fetched entry for file %s epoch %"PRIx64" key %s\n", + d->base_name, + d->unique_content_epoch, + file_id_str_buf(id, &idbuf)); + + return d; +} + +/* + * 132 is the sizeof an ndr-encoded struct share_mode_entry_buf. + * Reading/writing entries will immediately error out if this + * size differs (push/pull is done without allocs). + */ + +struct share_mode_entry_buf { + uint8_t buf[132]; +}; +#define SHARE_MODE_ENTRY_SIZE (sizeof(struct share_mode_entry_buf)) + +static bool share_mode_entry_put( + const struct share_mode_entry *e, + struct share_mode_entry_buf *dst) +{ + DATA_BLOB blob = { .data = dst->buf, .length = sizeof(dst->buf) }; + enum ndr_err_code ndr_err; + + if (DEBUGLEVEL>=10) { + DBG_DEBUG("share_mode_entry:\n"); + NDR_PRINT_DEBUG(share_mode_entry, discard_const_p(void, e)); + } + + ndr_err = ndr_push_struct_into_fixed_blob( + &blob, + e, + (ndr_push_flags_fn_t)ndr_push_share_mode_entry); + if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) { + DBG_WARNING("ndr_push_share_mode_entry failed: %s\n", + ndr_errstr(ndr_err)); + return false; + } + + return true; +} + +static bool share_mode_entry_get( + const uint8_t ptr[SHARE_MODE_ENTRY_SIZE], struct share_mode_entry *e) +{ + enum ndr_err_code ndr_err = NDR_ERR_SUCCESS; + DATA_BLOB blob = { + .data = discard_const_p(uint8_t, ptr), + .length = SHARE_MODE_ENTRY_SIZE, + }; + + ndr_err = ndr_pull_struct_blob_all_noalloc( + &blob, e, (ndr_pull_flags_fn_t)ndr_pull_share_mode_entry); + if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) { + DBG_WARNING("ndr_pull_share_mode_entry failed\n"); + return false; + } + return true; +} + +/* + * locking.tdb records consist of + * + * uint32_t share_mode_data_len + * uint8_t [share_mode_data] This is struct share_mode_data in NDR + * + * 0 [SHARE_MODE_ENTRY_SIZE] Sorted array of share modes, + * 1 [SHARE_MODE_ENTRY_SIZE] filling up the rest of the data in the + * 2 [SHARE_MODE_ENTRY_SIZE] g_lock.c maintained record in locking.tdb + */ + +struct locking_tdb_data { + const uint8_t *share_mode_data_buf; + size_t share_mode_data_len; + const uint8_t *share_entries; + size_t num_share_entries; +}; + +static bool locking_tdb_data_get( + struct locking_tdb_data *data, const uint8_t *buf, size_t buflen) +{ + uint32_t share_mode_data_len, share_entries_len; + + if (buflen == 0) { + *data = (struct locking_tdb_data) { 0 }; + return true; + } + if (buflen < sizeof(uint32_t)) { + return false; + } + + share_mode_data_len = PULL_LE_U32(buf, 0); + + buf += sizeof(uint32_t); + buflen -= sizeof(uint32_t); + + if (buflen < share_mode_data_len) { + return false; + } + + share_entries_len = buflen - share_mode_data_len; + + if ((share_entries_len % SHARE_MODE_ENTRY_SIZE) != 0) { + return false; + } + + *data = (struct locking_tdb_data) { + .share_mode_data_buf = buf, + .share_mode_data_len = share_mode_data_len, + .share_entries = buf + share_mode_data_len, + .num_share_entries = share_entries_len / SHARE_MODE_ENTRY_SIZE, + }; + + return true; +} + +struct locking_tdb_data_fetch_state { + TALLOC_CTX *mem_ctx; + uint8_t *data; + size_t datalen; +}; + +static void locking_tdb_data_fetch_fn( + struct server_id exclusive, + size_t num_shared, + const struct server_id *shared, + const uint8_t *data, + size_t datalen, + void *private_data) +{ + struct locking_tdb_data_fetch_state *state = private_data; + state->datalen = datalen; + state->data = talloc_memdup(state->mem_ctx, data, datalen); +} + +static NTSTATUS locking_tdb_data_fetch( + TDB_DATA key, TALLOC_CTX *mem_ctx, struct locking_tdb_data **ltdb) +{ + struct locking_tdb_data_fetch_state state = { 0 }; + struct locking_tdb_data *result = NULL; + NTSTATUS status; + bool ok; + + result = talloc_zero(mem_ctx, struct locking_tdb_data); + if (result == NULL) { + return NT_STATUS_NO_MEMORY; + } + state.mem_ctx = result; + + status = g_lock_dump(lock_ctx, key, locking_tdb_data_fetch_fn, &state); + + if (NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND)) { + /* + * Just return an empty record + */ + goto done; + } + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("g_lock_dump failed: %s\n", nt_errstr(status)); + return status; + } + if (state.datalen == 0) { + goto done; + } + + ok = locking_tdb_data_get(result, state.data, state.datalen); + if (!ok) { + DBG_DEBUG("locking_tdb_data_get failed for %zu bytes\n", + state.datalen); + TALLOC_FREE(result); + return NT_STATUS_INTERNAL_DB_CORRUPTION; + } + +done: + *ltdb = result; + return NT_STATUS_OK; +} + +static NTSTATUS locking_tdb_data_store( + TDB_DATA key, + const struct locking_tdb_data *ltdb, + const TDB_DATA *share_mode_dbufs, + size_t num_share_mode_dbufs) +{ + uint8_t share_mode_data_len_buf[4]; + TDB_DATA dbufs[num_share_mode_dbufs+3]; + NTSTATUS status; + + if ((ltdb->share_mode_data_len == 0) && + (ltdb->num_share_entries == 0) && + (num_share_mode_dbufs == 0)) { + /* + * Nothing to write + */ + status = g_lock_write_data(lock_ctx, key, NULL, 0); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("g_lock_writev_data() failed: %s\n", + nt_errstr(status)); + } + return status; + } + + PUSH_LE_U32(share_mode_data_len_buf, 0, ltdb->share_mode_data_len); + + dbufs[0] = (TDB_DATA) { + .dptr = share_mode_data_len_buf, + .dsize = sizeof(share_mode_data_len_buf), + }; + dbufs[1] = (TDB_DATA) { + .dptr = discard_const_p(uint8_t, ltdb->share_mode_data_buf), + .dsize = ltdb->share_mode_data_len, + }; + + if (ltdb->num_share_entries > SIZE_MAX/SHARE_MODE_ENTRY_SIZE) { + /* overflow */ + return NT_STATUS_BUFFER_OVERFLOW; + } + dbufs[2] = (TDB_DATA) { + .dptr = discard_const_p(uint8_t, ltdb->share_entries), + .dsize = ltdb->num_share_entries * SHARE_MODE_ENTRY_SIZE, + }; + + if (num_share_mode_dbufs != 0) { + memcpy(&dbufs[3], + share_mode_dbufs, + num_share_mode_dbufs * sizeof(TDB_DATA)); + } + + status = g_lock_writev_data(lock_ctx, key, dbufs, ARRAY_SIZE(dbufs)); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("g_lock_writev_data() failed: %s\n", + nt_errstr(status)); + } + return status; +} + +/******************************************************************* + Get all share mode entries for a dev/inode pair. +********************************************************************/ + +static struct share_mode_data *parse_share_modes( + TALLOC_CTX *mem_ctx, + struct file_id id, + const uint8_t *buf, + size_t buflen) +{ + struct share_mode_data *d; + enum ndr_err_code ndr_err; + DATA_BLOB blob; + + /* See if we already have a cached copy of this key. */ + d = share_mode_memcache_fetch(mem_ctx, id, buf, buflen); + if (d != NULL) { + return d; + } + + d = talloc(mem_ctx, struct share_mode_data); + if (d == NULL) { + DEBUG(0, ("talloc failed\n")); + goto fail; + } + + blob = (DATA_BLOB) { + .data = discard_const_p(uint8_t, buf), + .length = buflen, + }; + ndr_err = ndr_pull_struct_blob_all( + &blob, d, d, (ndr_pull_flags_fn_t)ndr_pull_share_mode_data); + if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) { + DBG_WARNING("ndr_pull_share_mode_data failed: %s\n", + ndr_errstr(ndr_err)); + goto fail; + } + + if (DEBUGLEVEL >= 10) { + DEBUG(10, ("parse_share_modes:\n")); + NDR_PRINT_DEBUG(share_mode_data, d); + } + + return d; +fail: + TALLOC_FREE(d); + return NULL; +} + +/******************************************************************* + If modified, store the share_mode_data back into the database. +********************************************************************/ + +static NTSTATUS share_mode_data_store( + struct share_mode_data *d, bool *have_share_entries) +{ + TDB_DATA key = locking_key(&d->id); + struct locking_tdb_data *ltdb = NULL; + DATA_BLOB blob = { 0 }; + NTSTATUS status; + + if (!d->modified) { + DBG_DEBUG("not modified\n"); + return NT_STATUS_OK; + } + + if (DEBUGLEVEL >= 10) { + DBG_DEBUG("\n"); + NDR_PRINT_DEBUG(share_mode_data, d); + } + + d->unique_content_epoch = generate_unique_u64(d->unique_content_epoch); + + status = locking_tdb_data_fetch(key, d, <db); + if (!NT_STATUS_IS_OK(status)) { + return status; + } + + if (ltdb->num_share_entries != 0) { + enum ndr_err_code ndr_err; + + ndr_err = ndr_push_struct_blob( + &blob, + ltdb, + d, + (ndr_push_flags_fn_t)ndr_push_share_mode_data); + if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) { + DBG_DEBUG("ndr_push_share_mode_data failed: %s\n", + ndr_errstr(ndr_err)); + TALLOC_FREE(ltdb); + return ndr_map_error2ntstatus(ndr_err); + } + + *have_share_entries = true; + } + + ltdb->share_mode_data_buf = blob.data; + ltdb->share_mode_data_len = blob.length; + + status = locking_tdb_data_store(key, ltdb, NULL, 0); + TALLOC_FREE(ltdb); + return status; +} + +/******************************************************************* + Allocate a new share_mode_data struct, mark it unmodified. + fresh is set to note that currently there is no database entry. +********************************************************************/ + +static struct share_mode_data *fresh_share_mode_lock( + TALLOC_CTX *mem_ctx, const char *servicepath, + const struct smb_filename *smb_fname, + const struct timespec *old_write_time) +{ + struct share_mode_data *d; + + if ((servicepath == NULL) || (smb_fname == NULL) || + (old_write_time == NULL)) { + return NULL; + } + + d = talloc_zero(mem_ctx, struct share_mode_data); + if (d == NULL) { + goto fail; + } + d->unique_content_epoch = generate_unique_u64(0); + + d->base_name = talloc_strdup(d, smb_fname->base_name); + if (d->base_name == NULL) { + goto fail; + } + if (smb_fname->stream_name != NULL) { + d->stream_name = talloc_strdup(d, smb_fname->stream_name); + if (d->stream_name == NULL) { + goto fail; + } + } + d->servicepath = talloc_strdup(d, servicepath); + if (d->servicepath == NULL) { + goto fail; + } + d->old_write_time = full_timespec_to_nt_time(old_write_time); + d->flags = SHARE_MODE_SHARE_DELETE | + SHARE_MODE_SHARE_WRITE | + SHARE_MODE_SHARE_READ; + d->modified = false; + d->fresh = true; + return d; +fail: + DEBUG(0, ("talloc failed\n")); + TALLOC_FREE(d); + return NULL; +} + +/* + * Key that's locked with g_lock + */ +static uint8_t share_mode_lock_key_data[sizeof(struct file_id)]; +static TDB_DATA share_mode_lock_key = { + .dptr = share_mode_lock_key_data, + .dsize = sizeof(share_mode_lock_key_data), +}; +static size_t share_mode_lock_key_refcount = 0; + +/* + * We can only ever have one share mode locked. Use a static + * share_mode_data pointer that is shared by multiple nested + * share_mode_lock structures, explicitly refcounted. + */ +static struct share_mode_data *static_share_mode_data = NULL; +static size_t static_share_mode_data_refcount = 0; + +/******************************************************************* + Either fetch a share mode from the database, or allocate a fresh + one if the record doesn't exist. +********************************************************************/ + +struct get_static_share_mode_data_state { + TALLOC_CTX *mem_ctx; + struct file_id id; + const char *servicepath; + const struct smb_filename *smb_fname; + const struct timespec *old_write_time; + NTSTATUS status; +}; + +static void get_static_share_mode_data_fn( + struct server_id exclusive, + size_t num_shared, + const struct server_id *shared, + const uint8_t *data, + size_t datalen, + void *private_data) +{ + struct get_static_share_mode_data_state *state = private_data; + struct share_mode_data *d = NULL; + struct locking_tdb_data ltdb = { 0 }; + + if (datalen != 0) { + bool ok; + + ok = locking_tdb_data_get(<db, data, datalen); + if (!ok) { + DBG_DEBUG("locking_tdb_data_get failed\n"); + state->status = NT_STATUS_INTERNAL_DB_CORRUPTION; + return; + } + } + + if (ltdb.share_mode_data_len == 0) { + if (state->smb_fname == NULL) { + state->status = NT_STATUS_NOT_FOUND; + return; + } + d = fresh_share_mode_lock( + state->mem_ctx, + state->servicepath, + state->smb_fname, + state->old_write_time); + if (d == NULL) { + state->status = NT_STATUS_NO_MEMORY; + return; + } + } else { + d = parse_share_modes( + lock_ctx, + state->id, + ltdb.share_mode_data_buf, + ltdb.share_mode_data_len); + if (d == NULL) { + state->status = NT_STATUS_INTERNAL_DB_CORRUPTION; + return; + } + } + + d->id = state->id; + static_share_mode_data = d; +} + +static NTSTATUS get_static_share_mode_data( + struct file_id id, + const char *servicepath, + const struct smb_filename *smb_fname, + const struct timespec *old_write_time) +{ + struct get_static_share_mode_data_state state = { + .mem_ctx = lock_ctx, + .id = id, + .servicepath = servicepath, + .smb_fname = smb_fname, + .old_write_time = old_write_time, + }; + NTSTATUS status; + + SMB_ASSERT(static_share_mode_data == NULL); + + status = g_lock_dump( + lock_ctx, + share_mode_lock_key, + get_static_share_mode_data_fn, + &state); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("g_lock_dump failed: %s\n", + nt_errstr(status)); + return status; + } + if (!NT_STATUS_IS_OK(state.status)) { + DBG_DEBUG("get_static_share_mode_data_fn failed: %s\n", + nt_errstr(state.status)); + return state.status; + } + + return NT_STATUS_OK; +} + +/******************************************************************* + Get a share_mode_lock, Reference counted to allow nested calls. +********************************************************************/ + +static int share_mode_lock_destructor(struct share_mode_lock *lck); + +struct share_mode_lock *get_share_mode_lock( + TALLOC_CTX *mem_ctx, + struct file_id id, + const char *servicepath, + const struct smb_filename *smb_fname, + const struct timespec *old_write_time) +{ + TDB_DATA key = locking_key(&id); + struct share_mode_lock *lck = NULL; + NTSTATUS status; + int cmp; + + lck = talloc(mem_ctx, struct share_mode_lock); + if (lck == NULL) { + DEBUG(1, ("talloc failed\n")); + return NULL; + } + + if (static_share_mode_data != NULL) { + if (!file_id_equal(&static_share_mode_data->id, &id)) { + struct file_id_buf existing; + struct file_id_buf requested; + + DBG_ERR("Can not lock two share modes " + "simultaneously: existing %s requested %s\n", + file_id_str_buf(static_share_mode_data->id, &existing), + file_id_str_buf(id, &requested)); + + smb_panic(__location__); + goto fail; + } + goto done; + } + + if (share_mode_lock_key_refcount == 0) { + status = g_lock_lock( + lock_ctx, + key, + G_LOCK_WRITE, + (struct timeval) { .tv_sec = 3600 }); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("g_lock_lock failed: %s\n", + nt_errstr(status)); + goto fail; + } + memcpy(share_mode_lock_key_data, key.dptr, key.dsize); + } + + cmp = tdb_data_cmp(share_mode_lock_key, key); + if (cmp != 0) { + DBG_WARNING("Can not lock two share modes simultaneously\n"); + smb_panic(__location__); + goto fail; + } + + SMB_ASSERT(share_mode_lock_key_refcount < SIZE_MAX); + share_mode_lock_key_refcount += 1; + + SMB_ASSERT(static_share_mode_data_refcount == 0); + + status = get_static_share_mode_data( + id, + servicepath, + smb_fname, + old_write_time); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("get_static_share_mode_data failed: %s\n", + nt_errstr(status)); + share_mode_lock_key_refcount -= 1; + goto fail; + } +done: + static_share_mode_data_refcount += 1; + lck->data = static_share_mode_data; + + talloc_set_destructor(lck, share_mode_lock_destructor); + + if (CHECK_DEBUGLVL(DBGLVL_DEBUG)) { + struct file_id_buf returned; + + DBG_DEBUG("Returning %s (data_refcount=%zu key_refcount=%zu)\n", + file_id_str_buf(id, &returned), + static_share_mode_data_refcount, + share_mode_lock_key_refcount); + } + + return lck; +fail: + TALLOC_FREE(lck); + if (share_mode_lock_key_refcount == 0) { + status = g_lock_unlock(lock_ctx, share_mode_lock_key); + if (!NT_STATUS_IS_OK(status)) { + DBG_ERR("g_lock_unlock failed: %s\n", + nt_errstr(status)); + } + } + return NULL; +} + +static int share_mode_lock_destructor(struct share_mode_lock *lck) +{ + bool have_share_entries = false; + NTSTATUS status; + + SMB_ASSERT(static_share_mode_data_refcount > 0); + static_share_mode_data_refcount -= 1; + + if (static_share_mode_data_refcount > 0) { + return 0; + } + + status = share_mode_data_store( + static_share_mode_data, &have_share_entries); + if (!NT_STATUS_IS_OK(status)) { + DBG_ERR("share_mode_data_store failed: %s\n", + nt_errstr(status)); + smb_panic("Could not store share mode data\n"); + } + + SMB_ASSERT(share_mode_lock_key_refcount > 0); + share_mode_lock_key_refcount -= 1; + + if (share_mode_lock_key_refcount == 0) { + status = g_lock_unlock(lock_ctx, share_mode_lock_key); + if (!NT_STATUS_IS_OK(status)) { + DBG_ERR("g_lock_unlock failed: %s\n", + nt_errstr(status)); + smb_panic("Could not unlock share mode\n"); + } + } + + if (have_share_entries) { + /* + * This is worth keeping. Without share modes, + * share_mode_data_store above has left nothing in the + * database. + */ + share_mode_memcache_store(static_share_mode_data); + static_share_mode_data = NULL; + } + + TALLOC_FREE(static_share_mode_data); + return 0; +} + +/******************************************************************* + Fetch a share mode where we know one MUST exist. This call reference + counts it internally to allow for nested lock fetches. +********************************************************************/ + +struct share_mode_lock *get_existing_share_mode_lock(TALLOC_CTX *mem_ctx, + const struct file_id id) +{ + return get_share_mode_lock(mem_ctx, id, NULL, NULL, NULL); +} + +struct share_mode_do_locked_state { + TDB_DATA key; + void (*fn)(const uint8_t *buf, + size_t buflen, + bool *modified_dependent, + void *private_data); + void *private_data; +}; + +static void share_mode_do_locked_fn( + struct server_id exclusive, + size_t num_shared, + const struct server_id *shared, + const uint8_t *data, + size_t datalen, + void *private_data) +{ + struct share_mode_do_locked_state *state = private_data; + bool modified_dependent = false; + struct locking_tdb_data ltdb = { 0 }; + bool ok; + + ok = locking_tdb_data_get( + <db, discard_const_p(uint8_t, data), datalen); + if (!ok) { + DBG_WARNING("locking_tdb_data_get failed\n"); + return; + } + + state->fn(ltdb.share_mode_data_buf, + ltdb.share_mode_data_len, + &modified_dependent, + state->private_data); + + if (modified_dependent) { + g_lock_wake_watchers(lock_ctx, state->key); + } +} + +NTSTATUS share_mode_do_locked( + struct file_id id, + void (*fn)(const uint8_t *buf, + size_t buflen, + bool *modified_dependent, + void *private_data), + void *private_data) +{ + TDB_DATA key = locking_key(&id); + size_t data_refcount, key_refcount; + struct share_mode_do_locked_state state = { + .key = key, .fn = fn, .private_data = private_data, + }; + NTSTATUS status; + + if (share_mode_lock_key_refcount == 0) { + status = g_lock_lock( + lock_ctx, + key, + G_LOCK_WRITE, + (struct timeval) { .tv_sec = 3600 }); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("g_lock_lock failed: %s\n", + nt_errstr(status)); + return status; + } + memcpy(share_mode_lock_key_data, key.dptr, key.dsize); + } + + SMB_ASSERT(share_mode_lock_key_refcount < SIZE_MAX); + share_mode_lock_key_refcount += 1; + + key_refcount = share_mode_lock_key_refcount; + data_refcount = static_share_mode_data_refcount; + + status = g_lock_dump( + lock_ctx, key, share_mode_do_locked_fn, &state); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("g_lock_dump failed: %s\n", + nt_errstr(status)); + } + + SMB_ASSERT(data_refcount == static_share_mode_data_refcount); + SMB_ASSERT(key_refcount == share_mode_lock_key_refcount); + share_mode_lock_key_refcount -= 1; + + if (share_mode_lock_key_refcount == 0) { + status = g_lock_unlock(lock_ctx, key); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("g_lock_unlock failed: %s\n", + nt_errstr(status)); + } + } + + return status; +} + +static void share_mode_wakeup_waiters_fn( + const uint8_t *buf, + size_t buflen, + bool *modified_dependent, + void *private_data) +{ + *modified_dependent = true; +} + +NTSTATUS share_mode_wakeup_waiters(struct file_id id) +{ + return share_mode_do_locked(id, share_mode_wakeup_waiters_fn, NULL); +} + +NTTIME share_mode_changed_write_time(struct share_mode_lock *lck) +{ + return lck->data->changed_write_time; +} + +const char *share_mode_servicepath(struct share_mode_lock *lck) +{ + return lck->data->servicepath; +} + +char *share_mode_filename(TALLOC_CTX *mem_ctx, struct share_mode_lock *lck) +{ + struct share_mode_data *d = lck->data; + bool has_stream = (d->stream_name != NULL); + char *fname = NULL; + + fname = talloc_asprintf( + mem_ctx, + "%s%s%s", + d->base_name, + has_stream ? ":" : "", + has_stream ? d->stream_name : ""); + return fname; +} + +char *share_mode_data_dump( + TALLOC_CTX *mem_ctx, struct share_mode_lock *lck) +{ + struct ndr_print *p = talloc(mem_ctx, struct ndr_print); + char *ret = NULL; + + if (p == NULL) { + return NULL; + } + + *p = (struct ndr_print) { + .print = ndr_print_string_helper, + .depth = 1, + .private_data = talloc_strdup(mem_ctx, ""), + }; + + if (p->private_data == NULL) { + TALLOC_FREE(p); + return NULL; + } + + ndr_print_share_mode_data(p, "SHARE_MODE_DATA", lck->data); + + ret = p->private_data; + + TALLOC_FREE(p); + + return ret; +} + +void share_mode_flags_get( + struct share_mode_lock *lck, + uint32_t *access_mask, + uint32_t *share_mode, + uint32_t *lease_type) +{ + uint16_t flags = lck->data->flags; + + if (access_mask != NULL) { + *access_mask = + ((flags & SHARE_MODE_ACCESS_READ) ? + FILE_READ_DATA : 0) | + ((flags & SHARE_MODE_ACCESS_WRITE) ? + FILE_WRITE_DATA : 0) | + ((flags & SHARE_MODE_ACCESS_DELETE) ? + DELETE_ACCESS : 0); + } + if (share_mode != NULL) { + *share_mode = + ((flags & SHARE_MODE_SHARE_READ) ? + FILE_SHARE_READ : 0) | + ((flags & SHARE_MODE_SHARE_WRITE) ? + FILE_SHARE_WRITE : 0) | + ((flags & SHARE_MODE_SHARE_DELETE) ? + FILE_SHARE_DELETE : 0); + } + if (lease_type != NULL) { + *lease_type = + ((flags & SHARE_MODE_LEASE_READ) ? + SMB2_LEASE_READ : 0) | + ((flags & SHARE_MODE_LEASE_WRITE) ? + SMB2_LEASE_WRITE : 0) | + ((flags & SHARE_MODE_LEASE_HANDLE) ? + SMB2_LEASE_HANDLE : 0); + } +} + +void share_mode_flags_set( + struct share_mode_lock *lck, + uint32_t access_mask, + uint32_t share_mode, + uint32_t lease_type, + bool *modified) +{ + struct share_mode_data *d = lck->data; + uint16_t flags = 0; + + flags |= (access_mask & (FILE_READ_DATA | FILE_EXECUTE)) ? + SHARE_MODE_ACCESS_READ : 0; + flags |= (access_mask & (FILE_WRITE_DATA | FILE_APPEND_DATA)) ? + SHARE_MODE_ACCESS_WRITE : 0; + flags |= (access_mask & (DELETE_ACCESS)) ? + SHARE_MODE_ACCESS_DELETE : 0; + + flags |= (share_mode & FILE_SHARE_READ) ? + SHARE_MODE_SHARE_READ : 0; + flags |= (share_mode & FILE_SHARE_WRITE) ? + SHARE_MODE_SHARE_WRITE : 0; + flags |= (share_mode & FILE_SHARE_DELETE) ? + SHARE_MODE_SHARE_DELETE : 0; + + flags |= (lease_type & SMB2_LEASE_READ) ? + SHARE_MODE_LEASE_READ : 0; + flags |= (lease_type & SMB2_LEASE_WRITE) ? + SHARE_MODE_LEASE_WRITE : 0; + flags |= (lease_type & SMB2_LEASE_HANDLE) ? + SHARE_MODE_LEASE_HANDLE : 0; + + if (d->flags == flags) { + return; + } + + if (modified != NULL) { + *modified = true; + } + d->flags = flags; + d->modified = true; +} + +struct share_mode_watch_state { + bool blockerdead; + struct server_id blocker; +}; + +static void share_mode_watch_done(struct tevent_req *subreq); + +struct tevent_req *share_mode_watch_send( + TALLOC_CTX *mem_ctx, + struct tevent_context *ev, + struct share_mode_lock *lck, + struct server_id blocker) +{ + TDB_DATA key = locking_key(&lck->data->id); + struct tevent_req *req = NULL, *subreq = NULL; + struct share_mode_watch_state *state = NULL; + + req = tevent_req_create( + mem_ctx, &state, struct share_mode_watch_state); + if (req == NULL) { + return NULL; + } + + subreq = g_lock_watch_data_send(state, ev, lock_ctx, key, blocker); + if (tevent_req_nomem(subreq, req)) { + return tevent_req_post(req, ev); + } + tevent_req_set_callback(subreq, share_mode_watch_done, req); + return req; +} + +static void share_mode_watch_done(struct tevent_req *subreq) +{ + struct tevent_req *req = tevent_req_callback_data( + subreq, struct tevent_req); + struct share_mode_watch_state *state = tevent_req_data( + req, struct share_mode_watch_state); + NTSTATUS status; + + status = g_lock_watch_data_recv( + subreq, &state->blockerdead, &state->blocker); + if (tevent_req_nterror(req, status)) { + return; + } + tevent_req_done(req); +} + +NTSTATUS share_mode_watch_recv( + struct tevent_req *req, bool *blockerdead, struct server_id *blocker) +{ + struct share_mode_watch_state *state = tevent_req_data( + req, struct share_mode_watch_state); + NTSTATUS status; + + if (tevent_req_is_nterror(req, &status)) { + return status; + } + if (blockerdead != NULL) { + *blockerdead = state->blockerdead; + } + if (blocker != NULL) { + *blocker = state->blocker; + } + return NT_STATUS_OK; +} + +struct fetch_share_mode_unlocked_state { + TALLOC_CTX *mem_ctx; + struct file_id id; + struct share_mode_lock *lck; +}; + +static void fetch_share_mode_unlocked_parser( + struct server_id exclusive, + size_t num_shared, + const struct server_id *shared, + const uint8_t *data, + size_t datalen, + void *private_data) +{ + struct fetch_share_mode_unlocked_state *state = private_data; + struct locking_tdb_data ltdb = { 0 }; + + if (datalen != 0) { + bool ok = locking_tdb_data_get(<db, data, datalen); + if (!ok) { + DBG_DEBUG("locking_tdb_data_get failed\n"); + return; + } + } + + if (ltdb.share_mode_data_len == 0) { + /* Likely a ctdb tombstone record, ignore it */ + return; + } + + state->lck = talloc(state->mem_ctx, struct share_mode_lock); + if (state->lck == NULL) { + DEBUG(0, ("talloc failed\n")); + return; + } + + state->lck->data = parse_share_modes( + state->lck, + state->id, + ltdb.share_mode_data_buf, + ltdb.share_mode_data_len); + if (state->lck->data == NULL) { + DBG_DEBUG("parse_share_modes failed\n"); + TALLOC_FREE(state->lck); + } +} + +/******************************************************************* + Get a share_mode_lock without locking the database or reference + counting. Used by smbstatus to display existing share modes. +********************************************************************/ + +struct share_mode_lock *fetch_share_mode_unlocked(TALLOC_CTX *mem_ctx, + struct file_id id) +{ + struct fetch_share_mode_unlocked_state state = { + .mem_ctx = mem_ctx, + .id = id, + }; + TDB_DATA key = locking_key(&id); + NTSTATUS status; + + status = g_lock_dump( + lock_ctx, key, fetch_share_mode_unlocked_parser, &state); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("g_lock_dump failed: %s\n", nt_errstr(status)); + return NULL; + } + return state.lck; +} + +struct fetch_share_mode_state { + struct file_id id; + struct share_mode_lock *lck; + NTSTATUS status; +}; + +static void fetch_share_mode_fn( + struct server_id exclusive, + size_t num_shared, + const struct server_id *shared, + const uint8_t *data, + size_t datalen, + void *private_data); +static void fetch_share_mode_done(struct tevent_req *subreq); + +/** + * @brief Get a share_mode_lock without locking or refcounting + * + * This can be used in a clustered Samba environment where the async dbwrap + * request is sent over a socket to the local ctdbd. If the send queue is full + * and the caller was issuing multiple async dbwrap requests in a loop, the + * caller knows it's probably time to stop sending requests for now and try + * again later. + * + * @param[in] mem_ctx The talloc memory context to use. + * + * @param[in] ev The event context to work on. + * + * @param[in] id The file id for the locking.tdb key + * + * @param[out] queued This boolean out parameter tells the caller whether the + * async request is blocked in a full send queue: + * + * false := request is dispatched + * + * true := send queue is full, request waiting to be + * dispatched + * + * @return The new async request, NULL on error. + **/ +struct tevent_req *fetch_share_mode_send(TALLOC_CTX *mem_ctx, + struct tevent_context *ev, + struct file_id id, + bool *queued) +{ + struct tevent_req *req = NULL, *subreq = NULL; + struct fetch_share_mode_state *state = NULL; + + *queued = false; + + req = tevent_req_create(mem_ctx, &state, + struct fetch_share_mode_state); + if (req == NULL) { + return NULL; + } + state->id = id; + + subreq = g_lock_dump_send( + state, + ev, + lock_ctx, + locking_key(&id), + fetch_share_mode_fn, + state); + if (tevent_req_nomem(subreq, req)) { + return tevent_req_post(req, ev); + } + tevent_req_set_callback(subreq, fetch_share_mode_done, req); + return req; +} + +static void fetch_share_mode_fn( + struct server_id exclusive, + size_t num_shared, + const struct server_id *shared, + const uint8_t *data, + size_t datalen, + void *private_data) +{ + struct fetch_share_mode_state *state = talloc_get_type_abort( + private_data, struct fetch_share_mode_state); + struct locking_tdb_data ltdb = { 0 }; + + if (datalen != 0) { + bool ok = locking_tdb_data_get(<db, data, datalen); + if (!ok) { + DBG_DEBUG("locking_tdb_data_get failed\n"); + return; + } + } + + if (ltdb.share_mode_data_len == 0) { + /* Likely a ctdb tombstone record, ignore it */ + return; + } + + state->lck = talloc(state, struct share_mode_lock); + if (state->lck == NULL) { + DBG_WARNING("talloc failed\n"); + state->status = NT_STATUS_NO_MEMORY; + return; + } + + state->lck->data = parse_share_modes( + state->lck, + state->id, + ltdb.share_mode_data_buf, + ltdb.share_mode_data_len); + if (state->lck->data == NULL) { + DBG_DEBUG("parse_share_modes failed\n"); + state->status = NT_STATUS_INTERNAL_DB_CORRUPTION; + TALLOC_FREE(state->lck); + return; + } +} + +static void fetch_share_mode_done(struct tevent_req *subreq) +{ + struct tevent_req *req = tevent_req_callback_data( + subreq, struct tevent_req); + struct fetch_share_mode_state *state = tevent_req_data( + req, struct fetch_share_mode_state); + NTSTATUS status; + + status = g_lock_dump_recv(subreq); + TALLOC_FREE(subreq); + if (tevent_req_nterror(req, status)) { + return; + } + if (tevent_req_nterror(req, state->status)) { + return; + } + tevent_req_done(req); +} + +NTSTATUS fetch_share_mode_recv(struct tevent_req *req, + TALLOC_CTX *mem_ctx, + struct share_mode_lock **_lck) +{ + struct fetch_share_mode_state *state = tevent_req_data( + req, struct fetch_share_mode_state); + struct share_mode_lock *lck = NULL; + + NTSTATUS status; + + if (tevent_req_is_nterror(req, &status)) { + tevent_req_received(req); + return status; + } + + if (state->lck == NULL) { + tevent_req_received(req); + return NT_STATUS_NOT_FOUND; + } + + lck = talloc_move(mem_ctx, &state->lck); + + if (DEBUGLEVEL >= 10) { + DBG_DEBUG("share_mode_data:\n"); + NDR_PRINT_DEBUG(share_mode_data, lck->data); + } + + *_lck = lck; + tevent_req_received(req); + return NT_STATUS_OK; +} + +struct share_mode_forall_state { + TDB_DATA key; + int (*fn)(struct file_id fid, + const struct share_mode_data *data, + void *private_data); + void *private_data; +}; + +static void share_mode_forall_dump_fn( + struct server_id exclusive, + size_t num_shared, + const struct server_id *shared, + const uint8_t *data, + size_t datalen, + void *private_data) +{ + struct share_mode_forall_state *state = private_data; + struct file_id fid; + struct locking_tdb_data ltdb = { 0 }; + bool ok; + struct share_mode_data *d; + + if (state->key.dsize != sizeof(fid)) { + DBG_DEBUG("Got invalid key length %zu\n", state->key.dsize); + return; + } + memcpy(&fid, state->key.dptr, sizeof(fid)); + + ok = locking_tdb_data_get(<db, data, datalen); + if (!ok) { + DBG_DEBUG("locking_tdb_data_get() failed\n"); + return; + } + + d = parse_share_modes( + talloc_tos(), + fid, + ltdb.share_mode_data_buf, + ltdb.share_mode_data_len); + if (d == NULL) { + DBG_DEBUG("parse_share_modes() failed\n"); + return; + } + + state->fn(fid, d, state->private_data); + TALLOC_FREE(d); +} + +static int share_mode_forall_fn(TDB_DATA key, void *private_data) +{ + struct share_mode_forall_state *state = private_data; + NTSTATUS status; + + state->key = key; + + status = g_lock_dump( + lock_ctx, key, share_mode_forall_dump_fn, private_data); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("g_lock_dump failed: %s\n", + nt_errstr(status)); + } + return 0; +} + +int share_mode_forall(int (*fn)(struct file_id fid, + const struct share_mode_data *data, + void *private_data), + void *private_data) +{ + struct share_mode_forall_state state = { + .fn = fn, + .private_data = private_data + }; + int ret; + + if (lock_ctx == NULL) { + return 0; + } + + ret = g_lock_locks( + lock_ctx, share_mode_forall_fn, &state); + if (ret < 0) { + DBG_DEBUG("g_lock_locks failed\n"); + } + return ret; +} + +struct share_entry_forall_state { + struct file_id fid; + const struct share_mode_data *data; + int (*fn)(struct file_id fid, + const struct share_mode_data *data, + const struct share_mode_entry *entry, + void *private_data); + void *private_data; + int ret; +}; + +static bool share_entry_traverse_walker( + struct share_mode_entry *e, + bool *modified, + void *private_data) +{ + struct share_entry_forall_state *state = private_data; + + state->ret = state->fn( + state->fid, state->data, e, state->private_data); + return (state->ret != 0); +} + +static int share_entry_traverse_fn(struct file_id fid, + const struct share_mode_data *data, + void *private_data) +{ + struct share_entry_forall_state *state = private_data; + struct share_mode_lock lck = { + .data = discard_const_p(struct share_mode_data, data) + }; + bool ok; + + state->fid = fid; + state->data = data; + + ok = share_mode_forall_entries( + &lck, share_entry_traverse_walker, state); + if (!ok) { + DBG_DEBUG("share_mode_forall_entries failed\n"); + return false; + } + + return state->ret; +} + +/******************************************************************* + Call the specified function on each entry under management by the + share mode system. +********************************************************************/ + +int share_entry_forall(int (*fn)(struct file_id fid, + const struct share_mode_data *data, + const struct share_mode_entry *entry, + void *private_data), + void *private_data) +{ + struct share_entry_forall_state state = { + .fn = fn, .private_data = private_data }; + + return share_mode_forall(share_entry_traverse_fn, &state); +} + +static int share_mode_entry_cmp( + struct server_id pid1, + uint64_t share_file_id1, + struct server_id pid2, + uint64_t share_file_id2) +{ + int cmp; + + cmp = server_id_cmp(&pid1, &pid2); + if (cmp != 0) { + return cmp; + } + if (share_file_id1 != share_file_id2) { + return (share_file_id1 < share_file_id2) ? -1 : 1; + } + return 0; +} + +static size_t share_mode_entry_find( + const uint8_t *data, + size_t num_share_modes, + struct server_id pid, + uint64_t share_file_id, + struct share_mode_entry *e, + bool *match) +{ + ssize_t left, right, middle; + + *match = false; + + if (num_share_modes == 0) { + return 0; + } + + left = 0; + right = (num_share_modes-1); + + while (left <= right) { + const uint8_t *middle_ptr = NULL; + int cmp; + bool ok; + + middle = left + ((right - left) / 2); + middle_ptr = data + middle * SHARE_MODE_ENTRY_SIZE; + + DBG_DEBUG("left=%zu, right=%zu, middle=%zu, middle_ptr=%p\n", + left, + right, + middle, + middle_ptr); + + ok = share_mode_entry_get(middle_ptr, e); + if (!ok) { + DBG_DEBUG("share_mode_entry_get failed\n"); + return 0; + } + + cmp = share_mode_entry_cmp( + e->pid, e->share_file_id, pid, share_file_id); + if (cmp == 0) { + *match = true; + return middle; + } + + if (cmp < 0) { + right = middle-1; + } else { + left = middle+1; + } + } + + return left; +} + +bool set_share_mode(struct share_mode_lock *lck, + struct files_struct *fsp, + uid_t uid, + uint64_t mid, + uint16_t op_type, + const struct smb2_lease_key *lease_key, + uint32_t share_access, + uint32_t access_mask) +{ + struct share_mode_data *d = lck->data; + TDB_DATA key = locking_key(&d->id); + struct server_id my_pid = messaging_server_id( + fsp->conn->sconn->msg_ctx); + struct locking_tdb_data *ltdb = NULL; + size_t idx; + struct share_mode_entry e = { .pid.pid = 0 }; + struct share_mode_entry_buf e_buf; + NTSTATUS status; + bool ok, found; + + TDB_DATA dbufs[3]; + size_t num_dbufs = 0; + + status = locking_tdb_data_fetch(key, talloc_tos(), <db); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("locking_tdb_data_fetch failed: %s\n", + nt_errstr(status)); + return false; + } + DBG_DEBUG("num_share_modes=%zu\n", ltdb->num_share_entries); + + idx = share_mode_entry_find( + ltdb->share_entries, + ltdb->num_share_entries, + my_pid, + fh_get_gen_id(fsp->fh), + &e, + &found); + if (found) { + DBG_WARNING("Found duplicate share mode\n"); + status = NT_STATUS_INTERNAL_DB_CORRUPTION; + goto done; + } + + e = (struct share_mode_entry) { + .pid = my_pid, + .share_access = share_access, + .private_options = fh_get_private_options(fsp->fh), + .access_mask = access_mask, + .op_mid = mid, + .op_type = op_type, + .time.tv_sec = fsp->open_time.tv_sec, + .time.tv_usec = fsp->open_time.tv_usec, + .share_file_id = fh_get_gen_id(fsp->fh), + .uid = (uint32_t)uid, + .flags = (fsp->posix_flags & FSP_POSIX_FLAGS_OPEN) ? + SHARE_MODE_FLAG_POSIX_OPEN : 0, + .name_hash = fsp->name_hash, + }; + + if (op_type == LEASE_OPLOCK) { + const struct GUID *client_guid = fsp_client_guid(fsp); + e.client_guid = *client_guid; + e.lease_key = *lease_key; + } + + ok = share_mode_entry_put(&e, &e_buf); + if (!ok) { + DBG_DEBUG("share_mode_entry_put failed\n"); + status = NT_STATUS_INTERNAL_ERROR; + goto done; + } + + DBG_DEBUG("idx=%zu, found=%d\n", idx, (int)found); + + if (idx > 0) { + dbufs[num_dbufs] = (TDB_DATA) { + .dptr = discard_const_p(uint8_t, ltdb->share_entries), + .dsize = idx * SHARE_MODE_ENTRY_SIZE, + }; + num_dbufs += 1; + } + + dbufs[num_dbufs] = (TDB_DATA) { + .dptr = e_buf.buf, .dsize = SHARE_MODE_ENTRY_SIZE, + }; + num_dbufs += 1; + + if (idx < ltdb->num_share_entries) { + size_t num_after_idx = (ltdb->num_share_entries-idx); + dbufs[num_dbufs] = (TDB_DATA) { + .dptr = discard_const_p(uint8_t, ltdb->share_entries) + + idx * SHARE_MODE_ENTRY_SIZE, + .dsize = num_after_idx * SHARE_MODE_ENTRY_SIZE, + }; + num_dbufs += 1; + } + + { + size_t i; + for (i=0; i<num_dbufs; i++) { + DBG_DEBUG("dbufs[%zu]=(%p, %zu)\n", + i, + dbufs[i].dptr, + dbufs[i].dsize); + } + } + + if (num_dbufs == 1) { + /* + * Storing a fresh record with just one share entry + */ + d->modified = true; + } + + /* + * If there was any existing data in + * ltdb->share_entries, it's now been + * moved and we've split it into: + * + * num_dbufs = 3 + * dbufs[0] -> old sorted data less than new_entry + * dbufs[1] -> new_share_mode_entry + * dbufs[2] -> old sorted_data greater than new entry. + * + * So the old data inside ltdb->share_entries is + * no longer valid. + * + * If we're storing a brand new entry the + * dbufs look like: + * + * num_dbufs = 1 + * dbufs[0] -> new_share_mode_entry + * + * Either way we must set ltdb->share_entries = NULL + * and ltdb->num_share_entries = 0 so that + * locking_tdb_data_store() doesn't use it to + * store any data. It's no longer there. + */ + + ltdb->share_entries = NULL; + ltdb->num_share_entries = 0; + + status = locking_tdb_data_store(key, ltdb, dbufs, num_dbufs); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("locking_tdb_data_store failed: %s\n", + nt_errstr(status)); + } +done: + TALLOC_FREE(ltdb); + return NT_STATUS_IS_OK(status); +} + +static bool share_mode_for_one_entry( + bool (*fn)(struct share_mode_entry *e, + bool *modified, + void *private_data), + void *private_data, + size_t *i, + uint8_t *data, + size_t *num_share_modes, + bool *writeback) +{ + DATA_BLOB blob = { + .data = data + (*i) * SHARE_MODE_ENTRY_SIZE, + .length = SHARE_MODE_ENTRY_SIZE, + }; + struct share_mode_entry e = {.pid.pid=0}; + enum ndr_err_code ndr_err = NDR_ERR_SUCCESS; + bool modified = false; + bool stop = false; + struct server_id e_pid; + uint64_t e_share_file_id; + + ndr_err = ndr_pull_struct_blob_all_noalloc( + &blob, + &e, + (ndr_pull_flags_fn_t)ndr_pull_share_mode_entry); + if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) { + DBG_WARNING("ndr_pull_share_mode_entry failed\n"); + *i += 1; + return false; + } + if (DEBUGLEVEL >= 10) { + DBG_DEBUG("entry[%zu]:\n", *i); + NDR_PRINT_DEBUG(share_mode_entry, &e); + } + + e_pid = e.pid; + e_share_file_id = e.share_file_id; + + stop = fn(&e, &modified, private_data); + + DBG_DEBUG("entry[%zu]: modified=%d, e.stale=%d\n", + *i, + (int)modified, + (int)e.stale); + + if (e.stale) { + if (DEBUGLEVEL>=10) { + DBG_DEBUG("share_mode_entry:\n"); + NDR_PRINT_DEBUG(share_mode_entry, &e); + } + + if (*i < *num_share_modes) { + memmove(blob.data, + blob.data + SHARE_MODE_ENTRY_SIZE, + (*num_share_modes - *i - 1) * + SHARE_MODE_ENTRY_SIZE); + } + *num_share_modes -= 1; + *writeback = true; + return stop; + } + + if (modified) { + if (DEBUGLEVEL>=10) { + DBG_DEBUG("share_mode_entry:\n"); + NDR_PRINT_DEBUG(share_mode_entry, &e); + } + + /* + * Make sure sorting order is kept intact + */ + SMB_ASSERT(server_id_equal(&e_pid, &e.pid)); + SMB_ASSERT(e_share_file_id == e.share_file_id); + + ndr_err = ndr_push_struct_into_fixed_blob( + &blob, + &e, + (ndr_push_flags_fn_t) + ndr_push_share_mode_entry); + if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) { + DBG_WARNING("ndr_push_share_mode_entry " + "failed: %s\n", + ndr_errstr(ndr_err)); + /* + * Not much we can do, just ignore it + */ + } + *i += 1; + *writeback = true; + return stop; + } + + if (stop) { + return true; + } + + *i += 1; + return false; +} + +bool share_mode_forall_entries( + struct share_mode_lock *lck, + bool (*fn)(struct share_mode_entry *e, + bool *modified, + void *private_data), + void *private_data) +{ + struct share_mode_data *d = lck->data; + TDB_DATA key = locking_key(&d->id); + struct locking_tdb_data *ltdb = NULL; + uint8_t *share_entries = NULL; + size_t num_share_entries; + bool writeback = false; + NTSTATUS status; + bool stop = false; + size_t i; + + status = locking_tdb_data_fetch(key, talloc_tos(), <db); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("locking_tdb_data_fetch failed: %s\n", + nt_errstr(status)); + return false; + } + DBG_DEBUG("num_share_modes=%zu\n", ltdb->num_share_entries); + + num_share_entries = ltdb->num_share_entries; + share_entries = discard_const_p(uint8_t, ltdb->share_entries); + + i = 0; + while (i<num_share_entries) { + stop = share_mode_for_one_entry( + fn, + private_data, + &i, + share_entries, + &num_share_entries, + &writeback); + if (stop) { + break; + } + } + + DBG_DEBUG("num_share_entries=%zu, writeback=%d\n", + num_share_entries, + (int)writeback); + + if (!writeback) { + return true; + } + + if ((ltdb->num_share_entries != 0 ) && (num_share_entries == 0)) { + /* + * This routine wiped all share entries, let + * share_mode_data_store() delete the record + */ + d->modified = true; + } + + ltdb->num_share_entries = num_share_entries; + ltdb->share_entries = share_entries; + + status = locking_tdb_data_store(key, ltdb, NULL, 0); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("locking_tdb_data_store failed: %s\n", + nt_errstr(status)); + return false; + } + + return true; +} + +struct share_mode_count_entries_state { + size_t num_share_modes; + NTSTATUS status; +}; + +static void share_mode_count_entries_fn( + struct server_id exclusive, + size_t num_shared, + const struct server_id *shared, + const uint8_t *data, + size_t datalen, + void *private_data) +{ + struct share_mode_count_entries_state *state = private_data; + struct locking_tdb_data ltdb = { 0 }; + bool ok; + + ok = locking_tdb_data_get(<db, data, datalen); + if (!ok) { + DBG_WARNING("locking_tdb_data_get failed for %zu\n", datalen); + state->status = NT_STATUS_INTERNAL_DB_CORRUPTION; + return; + } + state->num_share_modes = ltdb.num_share_entries; + state->status = NT_STATUS_OK; +} + +NTSTATUS share_mode_count_entries(struct file_id fid, size_t *num_share_modes) +{ + struct share_mode_count_entries_state state = { + .status = NT_STATUS_NOT_FOUND, + }; + NTSTATUS status; + + status = g_lock_dump( + lock_ctx, + locking_key(&fid), + share_mode_count_entries_fn, + &state); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("g_lock_dump failed: %s\n", + nt_errstr(status)); + return status; + } + if (!NT_STATUS_IS_OK(state.status)) { + DBG_DEBUG("share_mode_count_entries_fn failed: %s\n", + nt_errstr(state.status)); + return state.status; + } + + *num_share_modes = state.num_share_modes; + return NT_STATUS_OK; +} + +static bool share_mode_entry_do( + struct share_mode_lock *lck, + struct server_id pid, + uint64_t share_file_id, + void (*fn)(struct share_mode_entry *e, + size_t num_share_modes, + bool *modified, + void *private_data), + void *private_data) +{ + struct share_mode_data *d = lck->data; + TDB_DATA key = locking_key(&d->id); + struct locking_tdb_data *ltdb = NULL; + size_t idx; + bool found = false; + bool modified = false; + struct share_mode_entry e; + uint8_t *e_ptr = NULL; + NTSTATUS status; + bool ret = false; + + status = locking_tdb_data_fetch(key, talloc_tos(), <db); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("locking_tdb_data_fetch failed: %s\n", + nt_errstr(status)); + return false; + } + DBG_DEBUG("num_share_modes=%zu\n", ltdb->num_share_entries); + + idx = share_mode_entry_find( + ltdb->share_entries, + ltdb->num_share_entries, + pid, + share_file_id, + &e, + &found); + if (!found) { + DBG_WARNING("Did not find share mode entry for %"PRIu64"\n", + share_file_id); + goto done; + } + + if (DEBUGLEVEL>=10) { + DBG_DEBUG("entry[%zu]:\n", idx); + NDR_PRINT_DEBUG(share_mode_entry, &e); + } + + fn(&e, ltdb->num_share_entries, &modified, private_data); + + DBG_DEBUG("entry[%zu]: modified=%d, e.stale=%d\n", + idx, + (int)modified, + (int)e.stale); + + if (!e.stale && !modified) { + ret = true; + goto done; + } + + e_ptr = discard_const_p(uint8_t, ltdb->share_entries) + + idx * SHARE_MODE_ENTRY_SIZE; + + if (e.stale) { + /* + * Move the rest down one entry + */ + size_t behind = ltdb->num_share_entries - idx - 1; + if (behind != 0) { + memmove(e_ptr, + e_ptr + SHARE_MODE_ENTRY_SIZE, + behind * SHARE_MODE_ENTRY_SIZE); + } + ltdb->num_share_entries -= 1; + + if (ltdb->num_share_entries == 0) { + /* + * Tell share_mode_lock_destructor() to delete + * the whole record + */ + d->modified = true; + } + + if (DEBUGLEVEL>=10) { + DBG_DEBUG("share_mode_entry:\n"); + NDR_PRINT_DEBUG(share_mode_entry, &e); + } + } else { + struct share_mode_entry_buf buf; + bool ok; + + if (ltdb->num_share_entries != 1) { + /* + * Make sure the sorting order stays intact + */ + SMB_ASSERT(server_id_equal(&e.pid, &pid)); + SMB_ASSERT(e.share_file_id == share_file_id); + } + + ok = share_mode_entry_put(&e, &buf); + if (!ok) { + DBG_DEBUG("share_mode_entry_put failed\n"); + goto done; + } + memcpy(e_ptr, buf.buf, SHARE_MODE_ENTRY_SIZE); + } + + status = locking_tdb_data_store(key, ltdb, NULL, 0); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("locking_tdb_data_store failed: %s\n", + nt_errstr(status)); + goto done; + } + + ret = true; +done: + TALLOC_FREE(ltdb); + return ret; +} + +struct del_share_mode_state { + bool ok; +}; + +static void del_share_mode_fn( + struct share_mode_entry *e, + size_t num_share_modes, + bool *modified, + void *private_data) +{ + struct del_share_mode_state *state = private_data; + e->stale = true; + state->ok = true; +} + +bool del_share_mode(struct share_mode_lock *lck, files_struct *fsp) +{ + struct del_share_mode_state state = { .ok = false }; + bool ok; + + ok = share_mode_entry_do( + lck, + messaging_server_id(fsp->conn->sconn->msg_ctx), + fh_get_gen_id(fsp->fh), + del_share_mode_fn, + &state); + if (!ok) { + DBG_DEBUG("share_mode_entry_do failed\n"); + return false; + } + if (!state.ok) { + DBG_DEBUG("del_share_mode_fn failed\n"); + return false; + } + return true; +} + +struct remove_share_oplock_state { + bool ok; +}; + +static void remove_share_oplock_fn( + struct share_mode_entry *e, + size_t num_share_modes, + bool *modified, + void *private_data) +{ + struct remove_share_oplock_state *state = private_data; + + e->op_type = NO_OPLOCK; + *modified = true; + state->ok = true; +} + +bool remove_share_oplock(struct share_mode_lock *lck, files_struct *fsp) +{ + struct remove_share_oplock_state state = { .ok = false }; + bool ok; + + ok = share_mode_entry_do( + lck, + messaging_server_id(fsp->conn->sconn->msg_ctx), + fh_get_gen_id(fsp->fh), + remove_share_oplock_fn, + &state); + if (!ok) { + DBG_DEBUG("share_mode_entry_do failed\n"); + return false; + } + if (!state.ok) { + DBG_DEBUG("remove_share_oplock_fn failed\n"); + return false; + } + + if (fsp->oplock_type == LEASE_OPLOCK) { + remove_lease_if_stale( + lck, + fsp_client_guid(fsp), + &fsp->lease->lease.lease_key); + } + + share_mode_wakeup_waiters(fsp->file_id); + + return true; +} + +struct downgrade_share_oplock_state { + bool ok; +}; + +static void downgrade_share_oplock_fn( + struct share_mode_entry *e, + size_t num_share_modes, + bool *modified, + void *private_data) +{ + struct downgrade_share_oplock_state *state = private_data; + + e->op_type = LEVEL_II_OPLOCK; + *modified = true; + state->ok = true; +} + +bool downgrade_share_oplock(struct share_mode_lock *lck, files_struct *fsp) +{ + struct downgrade_share_oplock_state state = { .ok = false }; + bool ok; + + ok = share_mode_entry_do( + lck, + messaging_server_id(fsp->conn->sconn->msg_ctx), + fh_get_gen_id(fsp->fh), + downgrade_share_oplock_fn, + &state); + if (!ok) { + DBG_DEBUG("share_mode_entry_do failed\n"); + return false; + } + if (!state.ok) { + DBG_DEBUG("downgrade_share_oplock_fn failed\n"); + return false; + } + + lck->data->flags |= SHARE_MODE_LEASE_READ; + lck->data->modified = true; + + return true; +} + +bool mark_share_mode_disconnected(struct share_mode_lock *lck, + struct files_struct *fsp) +{ + struct server_id disconnected_pid = { .pid = 0 }; + bool ok; + + if (fsp->op == NULL) { + return false; + } + if (!fsp->op->global->durable) { + return false; + } + + server_id_set_disconnected(&disconnected_pid); + + ok = reset_share_mode_entry( + lck, + messaging_server_id(fsp->conn->sconn->msg_ctx), + fh_get_gen_id(fsp->fh), + disconnected_pid, + UINT64_MAX, + fsp->op->global->open_persistent_id); + + return ok; +} + +bool reset_share_mode_entry( + struct share_mode_lock *lck, + struct server_id old_pid, + uint64_t old_share_file_id, + struct server_id new_pid, + uint64_t new_mid, + uint64_t new_share_file_id) +{ + struct share_mode_data *d = lck->data; + TDB_DATA key = locking_key(&d->id); + struct locking_tdb_data *ltdb = NULL; + struct share_mode_entry e; + struct share_mode_entry_buf e_buf; + NTSTATUS status; + bool ret = false; + bool ok; + + status = locking_tdb_data_fetch(key, talloc_tos(), <db); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("locking_tdb_data_fetch failed: %s\n", + nt_errstr(status)); + return false; + } + + if (ltdb->num_share_entries != 1) { + DBG_DEBUG("num_share_modes=%zu\n", ltdb->num_share_entries); + goto done; + } + + ok = share_mode_entry_get(ltdb->share_entries, &e); + if (!ok) { + DBG_WARNING("share_mode_entry_get failed\n"); + goto done; + } + + ret = share_mode_entry_cmp( + old_pid, old_share_file_id, e.pid, e.share_file_id); + if (ret != 0) { + struct server_id_buf tmp1, tmp2; + DBG_WARNING("Expected pid=%s, file_id=%"PRIu64", " + "got pid=%s, file_id=%"PRIu64"\n", + server_id_str_buf(old_pid, &tmp1), + old_share_file_id, + server_id_str_buf(e.pid, &tmp2), + e.share_file_id); + goto done; + } + + e.pid = new_pid; + if (new_mid != UINT64_MAX) { + e.op_mid = new_mid; + } + e.share_file_id = new_share_file_id; + + ok = share_mode_entry_put(&e, &e_buf); + if (!ok) { + DBG_WARNING("share_mode_entry_put failed\n"); + goto done; + } + + ltdb->share_entries = e_buf.buf; + + status = locking_tdb_data_store(key, ltdb, NULL, 0); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("locking_tdb_data_store failed: %s\n", + nt_errstr(status)); + goto done; + } + + d->modified = true; + ret = true; +done: + TALLOC_FREE(ltdb); + return true; +} diff --git a/source3/locking/share_mode_lock.h b/source3/locking/share_mode_lock.h new file mode 100644 index 0000000..4f47b6f --- /dev/null +++ b/source3/locking/share_mode_lock.h @@ -0,0 +1,140 @@ +/* + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 3 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program. If not, see <http://www.gnu.org/licenses/>. + */ + +#ifndef __LOCKING_SHARE_MODE_LOCK_H__ +#define __LOCKING_SHARE_MODE_LOCK_H__ + +#include "replace.h" +#include <tevent.h> +#include "librpc/gen_ndr/file_id.h" +#include "lib/util/time.h" + +struct share_mode_data; +struct share_mode_lock; +struct share_mode_entry; +struct smb_filename; +struct files_struct; +struct smb2_lease_key; + +bool locking_init(void); +bool locking_init_readonly(void); +bool locking_end(void); + +struct share_mode_lock *get_share_mode_lock( + TALLOC_CTX *mem_ctx, + struct file_id id, + const char *servicepath, + const struct smb_filename *smb_fname, + const struct timespec *old_write_time); +struct share_mode_lock *get_existing_share_mode_lock(TALLOC_CTX *mem_ctx, + struct file_id id); + +bool del_share_mode(struct share_mode_lock *lck, + struct files_struct *fsp); +bool downgrade_share_oplock(struct share_mode_lock *lck, + struct files_struct *fsp); +bool remove_share_oplock(struct share_mode_lock *lck, + struct files_struct *fsp); +bool file_has_read_lease(struct files_struct *fsp); + +bool set_share_mode( + struct share_mode_lock *lck, + struct files_struct *fsp, + uid_t uid, + uint64_t mid, + uint16_t op_type, + const struct smb2_lease_key *lease_key, + uint32_t share_access, + uint32_t access_mask); +bool reset_share_mode_entry( + struct share_mode_lock *lck, + struct server_id old_pid, + uint64_t old_share_file_id, + struct server_id new_pid, + uint64_t new_mid, + uint64_t new_share_file_id); + +bool mark_share_mode_disconnected( + struct share_mode_lock *lck, struct files_struct *fsp); + +struct share_mode_lock *fetch_share_mode_unlocked( + TALLOC_CTX *mem_ctx, + struct file_id id); + +struct tevent_req *fetch_share_mode_send( + TALLOC_CTX *mem_ctx, + struct tevent_context *ev, + struct file_id id, + bool *queued); +NTSTATUS fetch_share_mode_recv( + struct tevent_req *req, + TALLOC_CTX *mem_ctx, + struct share_mode_lock **_lck); + +int share_entry_forall( + int (*fn)(struct file_id fid, + const struct share_mode_data *data, + const struct share_mode_entry *entry, + void *private_data), + void *private_data); + +NTSTATUS share_mode_count_entries(struct file_id fid, size_t *num_share_modes); +NTSTATUS share_mode_do_locked( + struct file_id id, + void (*fn)(const uint8_t *buf, + size_t buflen, + bool *modified_dependent, + void *private_data), + void *private_data); +int share_mode_forall( + int (*fn)(struct file_id fid, + const struct share_mode_data *data, + void *private_data), + void *private_data); +bool share_mode_forall_entries( + struct share_mode_lock *lck, + bool (*fn)(struct share_mode_entry *e, + bool *modified, + void *private_data), + void *private_data); + +NTTIME share_mode_changed_write_time(struct share_mode_lock *lck); +const char *share_mode_servicepath(struct share_mode_lock *lck); +char *share_mode_filename(TALLOC_CTX *mem_ctx, struct share_mode_lock *lck); +char *share_mode_data_dump( + TALLOC_CTX *mem_ctx, struct share_mode_lock *lck); + +void share_mode_flags_get( + struct share_mode_lock *lck, + uint32_t *access_mask, + uint32_t *share_mode, + uint32_t *lease_type); +void share_mode_flags_set( + struct share_mode_lock *lck, + uint32_t access_mask, + uint32_t share_mode, + uint32_t lease_type, + bool *modified); + +struct tevent_req *share_mode_watch_send( + TALLOC_CTX *mem_ctx, + struct tevent_context *ev, + struct share_mode_lock *lck, + struct server_id blocker); +NTSTATUS share_mode_watch_recv( + struct tevent_req *req, bool *blockerdead, struct server_id *blocker); +NTSTATUS share_mode_wakeup_waiters(struct file_id id); + +#endif diff --git a/source3/locking/share_mode_lock_private.h b/source3/locking/share_mode_lock_private.h new file mode 100644 index 0000000..14a3c1b --- /dev/null +++ b/source3/locking/share_mode_lock_private.h @@ -0,0 +1,24 @@ +/* + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 3 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program. If not, see <http://www.gnu.org/licenses/>. + */ + +#ifndef __LOCKING_SHARE_MODE_LOCK_PRIVATE_H__ +#define __LOCKING_SHARE_MODE_LOCK_PRIVATE_H__ + +struct share_mode_data; +struct share_mode_lock { + struct share_mode_data *data; +}; + +#endif |