1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
4 #include <linux/file.h>
5 #include <linux/namei.h>
6 #include <linux/random.h>
9 #include "mds_client.h"
10 #include <linux/ceph/pagelist.h>
12 static u64 lock_secret;
13 static int ceph_lock_wait_for_completion(struct ceph_mds_client *mdsc,
14 struct ceph_mds_request *req);
16 static inline u64 secure_addr(void *addr)
18 u64 v = lock_secret ^ (u64)(unsigned long)addr;
20 * Set the most significant bit, so that MDS knows the 'owner'
21 * is sufficient to identify the owner of lock. (old code uses
22 * both 'owner' and 'pid')
28 void __init ceph_flock_init(void)
30 get_random_bytes(&lock_secret, sizeof(lock_secret));
33 static void ceph_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
35 struct inode *inode = file_inode(src->fl_file);
36 atomic_inc(&ceph_inode(inode)->i_filelock_ref);
39 static void ceph_fl_release_lock(struct file_lock *fl)
41 struct inode *inode = file_inode(fl->fl_file);
42 atomic_dec(&ceph_inode(inode)->i_filelock_ref);
45 static const struct file_lock_operations ceph_fl_lock_ops = {
46 .fl_copy_lock = ceph_fl_copy_lock,
47 .fl_release_private = ceph_fl_release_lock,
51 * Implement fcntl and flock locking functions.
53 static int ceph_lock_message(u8 lock_type, u16 operation, struct inode *inode,
54 int cmd, u8 wait, struct file_lock *fl)
56 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
57 struct ceph_mds_request *req;
62 if (operation == CEPH_MDS_OP_SETFILELOCK) {
64 * increasing i_filelock_ref closes race window between
65 * handling request reply and adding file_lock struct to
66 * inode. Otherwise, auth caps may get trimmed in the
67 * window. Caller function will decrease the counter.
69 fl->fl_ops = &ceph_fl_lock_ops;
70 atomic_inc(&ceph_inode(inode)->i_filelock_ref);
73 if (operation != CEPH_MDS_OP_SETFILELOCK || cmd == CEPH_LOCK_UNLOCK)
76 req = ceph_mdsc_create_request(mdsc, operation, USE_AUTH_MDS);
83 /* mds requires start and length rather than start and end */
84 if (LLONG_MAX == fl->fl_end)
87 length = fl->fl_end - fl->fl_start + 1;
89 owner = secure_addr(fl->fl_owner);
91 dout("ceph_lock_message: rule: %d, op: %d, owner: %llx, pid: %llu, "
92 "start: %llu, length: %llu, wait: %d, type: %d", (int)lock_type,
93 (int)operation, owner, (u64)fl->fl_pid, fl->fl_start, length,
96 req->r_args.filelock_change.rule = lock_type;
97 req->r_args.filelock_change.type = cmd;
98 req->r_args.filelock_change.owner = cpu_to_le64(owner);
99 req->r_args.filelock_change.pid = cpu_to_le64((u64)fl->fl_pid);
100 req->r_args.filelock_change.start = cpu_to_le64(fl->fl_start);
101 req->r_args.filelock_change.length = cpu_to_le64(length);
102 req->r_args.filelock_change.wait = wait;
105 req->r_wait_for_completion = ceph_lock_wait_for_completion;
107 err = ceph_mdsc_do_request(mdsc, inode, req);
109 if (operation == CEPH_MDS_OP_GETFILELOCK) {
110 fl->fl_pid = -le64_to_cpu(req->r_reply_info.filelock_reply->pid);
111 if (CEPH_LOCK_SHARED == req->r_reply_info.filelock_reply->type)
112 fl->fl_type = F_RDLCK;
113 else if (CEPH_LOCK_EXCL == req->r_reply_info.filelock_reply->type)
114 fl->fl_type = F_WRLCK;
116 fl->fl_type = F_UNLCK;
118 fl->fl_start = le64_to_cpu(req->r_reply_info.filelock_reply->start);
119 length = le64_to_cpu(req->r_reply_info.filelock_reply->start) +
120 le64_to_cpu(req->r_reply_info.filelock_reply->length);
122 fl->fl_end = length -1;
127 ceph_mdsc_put_request(req);
128 dout("ceph_lock_message: rule: %d, op: %d, pid: %llu, start: %llu, "
129 "length: %llu, wait: %d, type: %d, err code %d", (int)lock_type,
130 (int)operation, (u64)fl->fl_pid, fl->fl_start,
131 length, wait, fl->fl_type, err);
135 static int ceph_lock_wait_for_completion(struct ceph_mds_client *mdsc,
136 struct ceph_mds_request *req)
138 struct ceph_mds_request *intr_req;
139 struct inode *inode = req->r_inode;
142 BUG_ON(req->r_op != CEPH_MDS_OP_SETFILELOCK);
143 if (req->r_args.filelock_change.rule == CEPH_LOCK_FCNTL)
144 lock_type = CEPH_LOCK_FCNTL_INTR;
145 else if (req->r_args.filelock_change.rule == CEPH_LOCK_FLOCK)
146 lock_type = CEPH_LOCK_FLOCK_INTR;
149 BUG_ON(req->r_args.filelock_change.type == CEPH_LOCK_UNLOCK);
151 err = wait_for_completion_interruptible(&req->r_completion);
155 dout("ceph_lock_wait_for_completion: request %llu was interrupted\n",
158 mutex_lock(&mdsc->mutex);
159 if (test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
163 * ensure we aren't running concurrently with
164 * ceph_fill_trace or ceph_readdir_prepopulate, which
165 * rely on locks (dir mutex) held by our caller.
167 mutex_lock(&req->r_fill_mutex);
169 set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
170 mutex_unlock(&req->r_fill_mutex);
172 if (!req->r_session) {
173 // haven't sent the request
177 mutex_unlock(&mdsc->mutex);
181 intr_req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETFILELOCK,
183 if (IS_ERR(intr_req))
184 return PTR_ERR(intr_req);
186 intr_req->r_inode = inode;
188 intr_req->r_num_caps = 1;
190 intr_req->r_args.filelock_change = req->r_args.filelock_change;
191 intr_req->r_args.filelock_change.rule = lock_type;
192 intr_req->r_args.filelock_change.type = CEPH_LOCK_UNLOCK;
194 err = ceph_mdsc_do_request(mdsc, inode, intr_req);
195 ceph_mdsc_put_request(intr_req);
197 if (err && err != -ERESTARTSYS)
200 wait_for_completion_killable(&req->r_safe_completion);
205 * Attempt to set an fcntl lock.
206 * For now, this just goes away to the server. Later it may be more awesome.
208 int ceph_lock(struct file *file, int cmd, struct file_lock *fl)
210 struct inode *inode = file_inode(file);
212 u16 op = CEPH_MDS_OP_SETFILELOCK;
216 if (!(fl->fl_flags & FL_POSIX))
218 /* No mandatory locks */
219 if (__mandatory_lock(file->f_mapping->host) && fl->fl_type != F_UNLCK)
222 dout("ceph_lock, fl_owner: %p", fl->fl_owner);
224 /* set wait bit as appropriate, then make command as Ceph expects it*/
226 op = CEPH_MDS_OP_GETFILELOCK;
227 else if (IS_SETLKW(cmd))
230 if (op == CEPH_MDS_OP_SETFILELOCK) {
232 * increasing i_filelock_ref closes race window between
233 * handling request reply and adding file_lock struct to
234 * inode. Otherwise, i_auth_cap may get trimmed in the
235 * window. Caller function will decrease the counter.
237 fl->fl_ops = &ceph_fl_lock_ops;
238 atomic_inc(&ceph_inode(inode)->i_filelock_ref);
241 if (F_RDLCK == fl->fl_type)
242 lock_cmd = CEPH_LOCK_SHARED;
243 else if (F_WRLCK == fl->fl_type)
244 lock_cmd = CEPH_LOCK_EXCL;
246 lock_cmd = CEPH_LOCK_UNLOCK;
248 err = ceph_lock_message(CEPH_LOCK_FCNTL, op, inode, lock_cmd, wait, fl);
250 if (op != CEPH_MDS_OP_GETFILELOCK) {
251 dout("mds locked, locking locally");
252 err = posix_lock_file(file, fl, NULL);
253 if (err && (CEPH_MDS_OP_SETFILELOCK == op)) {
254 /* undo! This should only happen if
255 * the kernel detects local
257 ceph_lock_message(CEPH_LOCK_FCNTL, op, inode,
258 CEPH_LOCK_UNLOCK, 0, fl);
259 dout("got %d on posix_lock_file, undid lock",
267 int ceph_flock(struct file *file, int cmd, struct file_lock *fl)
269 struct inode *inode = file_inode(file);
274 if (!(fl->fl_flags & FL_FLOCK))
276 /* No mandatory locks */
277 if (fl->fl_type & LOCK_MAND)
280 dout("ceph_flock, fl_file: %p", fl->fl_file);
282 /* see comment in ceph_lock */
283 fl->fl_ops = &ceph_fl_lock_ops;
284 atomic_inc(&ceph_inode(inode)->i_filelock_ref);
289 if (F_RDLCK == fl->fl_type)
290 lock_cmd = CEPH_LOCK_SHARED;
291 else if (F_WRLCK == fl->fl_type)
292 lock_cmd = CEPH_LOCK_EXCL;
294 lock_cmd = CEPH_LOCK_UNLOCK;
296 err = ceph_lock_message(CEPH_LOCK_FLOCK, CEPH_MDS_OP_SETFILELOCK,
297 inode, lock_cmd, wait, fl);
299 err = locks_lock_file_wait(file, fl);
301 ceph_lock_message(CEPH_LOCK_FLOCK,
302 CEPH_MDS_OP_SETFILELOCK,
303 inode, CEPH_LOCK_UNLOCK, 0, fl);
304 dout("got %d on locks_lock_file_wait, undid lock", err);
311 * Fills in the passed counter variables, so you can prepare pagelist metadata
312 * before calling ceph_encode_locks.
314 void ceph_count_locks(struct inode *inode, int *fcntl_count, int *flock_count)
316 struct file_lock *lock;
317 struct file_lock_context *ctx;
322 ctx = inode->i_flctx;
324 spin_lock(&ctx->flc_lock);
325 list_for_each_entry(lock, &ctx->flc_posix, fl_list)
327 list_for_each_entry(lock, &ctx->flc_flock, fl_list)
329 spin_unlock(&ctx->flc_lock);
331 dout("counted %d flock locks and %d fcntl locks",
332 *flock_count, *fcntl_count);
336 * Given a pointer to a lock, convert it to a ceph filelock
338 static int lock_to_ceph_filelock(struct file_lock *lock,
339 struct ceph_filelock *cephlock)
342 cephlock->start = cpu_to_le64(lock->fl_start);
343 cephlock->length = cpu_to_le64(lock->fl_end - lock->fl_start + 1);
344 cephlock->client = cpu_to_le64(0);
345 cephlock->pid = cpu_to_le64((u64)lock->fl_pid);
346 cephlock->owner = cpu_to_le64(secure_addr(lock->fl_owner));
348 switch (lock->fl_type) {
350 cephlock->type = CEPH_LOCK_SHARED;
353 cephlock->type = CEPH_LOCK_EXCL;
356 cephlock->type = CEPH_LOCK_UNLOCK;
359 dout("Have unknown lock type %d", lock->fl_type);
367 * Encode the flock and fcntl locks for the given inode into the ceph_filelock
368 * array. Must be called with inode->i_lock already held.
369 * If we encounter more of a specific lock type than expected, return -ENOSPC.
371 int ceph_encode_locks_to_buffer(struct inode *inode,
372 struct ceph_filelock *flocks,
373 int num_fcntl_locks, int num_flock_locks)
375 struct file_lock *lock;
376 struct file_lock_context *ctx = inode->i_flctx;
382 dout("encoding %d flock and %d fcntl locks", num_flock_locks,
388 spin_lock(&ctx->flc_lock);
389 list_for_each_entry(lock, &ctx->flc_posix, fl_list) {
391 if (seen_fcntl > num_fcntl_locks) {
395 err = lock_to_ceph_filelock(lock, &flocks[l]);
400 list_for_each_entry(lock, &ctx->flc_flock, fl_list) {
402 if (seen_flock > num_flock_locks) {
406 err = lock_to_ceph_filelock(lock, &flocks[l]);
412 spin_unlock(&ctx->flc_lock);
417 * Copy the encoded flock and fcntl locks into the pagelist.
418 * Format is: #fcntl locks, sequential fcntl locks, #flock locks,
419 * sequential flock locks.
420 * Returns zero on success.
422 int ceph_locks_to_pagelist(struct ceph_filelock *flocks,
423 struct ceph_pagelist *pagelist,
424 int num_fcntl_locks, int num_flock_locks)
429 nlocks = cpu_to_le32(num_fcntl_locks);
430 err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
434 if (num_fcntl_locks > 0) {
435 err = ceph_pagelist_append(pagelist, flocks,
436 num_fcntl_locks * sizeof(*flocks));
441 nlocks = cpu_to_le32(num_flock_locks);
442 err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
446 if (num_flock_locks > 0) {
447 err = ceph_pagelist_append(pagelist, &flocks[num_fcntl_locks],
448 num_flock_locks * sizeof(*flocks));