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 struct ceph_inode_info *ci = ceph_inode(inode);
43 if (atomic_dec_and_test(&ci->i_filelock_ref)) {
44 /* clear error when all locks are released */
45 spin_lock(&ci->i_ceph_lock);
46 ci->i_ceph_flags &= ~CEPH_I_ERROR_FILELOCK;
47 spin_unlock(&ci->i_ceph_lock);
51 static const struct file_lock_operations ceph_fl_lock_ops = {
52 .fl_copy_lock = ceph_fl_copy_lock,
53 .fl_release_private = ceph_fl_release_lock,
57 * Implement fcntl and flock locking functions.
59 static int ceph_lock_message(u8 lock_type, u16 operation, struct inode *inode,
60 int cmd, u8 wait, struct file_lock *fl)
62 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
63 struct ceph_mds_request *req;
68 if (operation == CEPH_MDS_OP_SETFILELOCK) {
70 * increasing i_filelock_ref closes race window between
71 * handling request reply and adding file_lock struct to
72 * inode. Otherwise, auth caps may get trimmed in the
73 * window. Caller function will decrease the counter.
75 fl->fl_ops = &ceph_fl_lock_ops;
76 atomic_inc(&ceph_inode(inode)->i_filelock_ref);
79 if (operation != CEPH_MDS_OP_SETFILELOCK || cmd == CEPH_LOCK_UNLOCK)
82 req = ceph_mdsc_create_request(mdsc, operation, USE_AUTH_MDS);
89 /* mds requires start and length rather than start and end */
90 if (LLONG_MAX == fl->fl_end)
93 length = fl->fl_end - fl->fl_start + 1;
95 owner = secure_addr(fl->fl_owner);
97 dout("ceph_lock_message: rule: %d, op: %d, owner: %llx, pid: %llu, "
98 "start: %llu, length: %llu, wait: %d, type: %d\n", (int)lock_type,
99 (int)operation, owner, (u64)fl->fl_pid, fl->fl_start, length,
102 req->r_args.filelock_change.rule = lock_type;
103 req->r_args.filelock_change.type = cmd;
104 req->r_args.filelock_change.owner = cpu_to_le64(owner);
105 req->r_args.filelock_change.pid = cpu_to_le64((u64)fl->fl_pid);
106 req->r_args.filelock_change.start = cpu_to_le64(fl->fl_start);
107 req->r_args.filelock_change.length = cpu_to_le64(length);
108 req->r_args.filelock_change.wait = wait;
111 req->r_wait_for_completion = ceph_lock_wait_for_completion;
113 err = ceph_mdsc_do_request(mdsc, inode, req);
114 if (!err && operation == CEPH_MDS_OP_GETFILELOCK) {
115 fl->fl_pid = -le64_to_cpu(req->r_reply_info.filelock_reply->pid);
116 if (CEPH_LOCK_SHARED == req->r_reply_info.filelock_reply->type)
117 fl->fl_type = F_RDLCK;
118 else if (CEPH_LOCK_EXCL == req->r_reply_info.filelock_reply->type)
119 fl->fl_type = F_WRLCK;
121 fl->fl_type = F_UNLCK;
123 fl->fl_start = le64_to_cpu(req->r_reply_info.filelock_reply->start);
124 length = le64_to_cpu(req->r_reply_info.filelock_reply->start) +
125 le64_to_cpu(req->r_reply_info.filelock_reply->length);
127 fl->fl_end = length -1;
132 ceph_mdsc_put_request(req);
133 dout("ceph_lock_message: rule: %d, op: %d, pid: %llu, start: %llu, "
134 "length: %llu, wait: %d, type: %d, err code %d\n", (int)lock_type,
135 (int)operation, (u64)fl->fl_pid, fl->fl_start,
136 length, wait, fl->fl_type, err);
140 static int ceph_lock_wait_for_completion(struct ceph_mds_client *mdsc,
141 struct ceph_mds_request *req)
143 struct ceph_mds_request *intr_req;
144 struct inode *inode = req->r_inode;
147 BUG_ON(req->r_op != CEPH_MDS_OP_SETFILELOCK);
148 if (req->r_args.filelock_change.rule == CEPH_LOCK_FCNTL)
149 lock_type = CEPH_LOCK_FCNTL_INTR;
150 else if (req->r_args.filelock_change.rule == CEPH_LOCK_FLOCK)
151 lock_type = CEPH_LOCK_FLOCK_INTR;
154 BUG_ON(req->r_args.filelock_change.type == CEPH_LOCK_UNLOCK);
156 err = wait_for_completion_interruptible(&req->r_completion);
160 dout("ceph_lock_wait_for_completion: request %llu was interrupted\n",
163 mutex_lock(&mdsc->mutex);
164 if (test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
168 * ensure we aren't running concurrently with
169 * ceph_fill_trace or ceph_readdir_prepopulate, which
170 * rely on locks (dir mutex) held by our caller.
172 mutex_lock(&req->r_fill_mutex);
174 set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
175 mutex_unlock(&req->r_fill_mutex);
177 if (!req->r_session) {
178 // haven't sent the request
182 mutex_unlock(&mdsc->mutex);
186 intr_req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETFILELOCK,
188 if (IS_ERR(intr_req))
189 return PTR_ERR(intr_req);
191 intr_req->r_inode = inode;
193 intr_req->r_num_caps = 1;
195 intr_req->r_args.filelock_change = req->r_args.filelock_change;
196 intr_req->r_args.filelock_change.rule = lock_type;
197 intr_req->r_args.filelock_change.type = CEPH_LOCK_UNLOCK;
199 err = ceph_mdsc_do_request(mdsc, inode, intr_req);
200 ceph_mdsc_put_request(intr_req);
202 if (err && err != -ERESTARTSYS)
205 wait_for_completion_killable(&req->r_safe_completion);
210 * Attempt to set an fcntl lock.
211 * For now, this just goes away to the server. Later it may be more awesome.
213 int ceph_lock(struct file *file, int cmd, struct file_lock *fl)
215 struct inode *inode = file_inode(file);
216 struct ceph_inode_info *ci = ceph_inode(inode);
218 u16 op = CEPH_MDS_OP_SETFILELOCK;
222 if (!(fl->fl_flags & FL_POSIX))
224 /* No mandatory locks */
225 if (__mandatory_lock(file->f_mapping->host) && fl->fl_type != F_UNLCK)
228 dout("ceph_lock, fl_owner: %p\n", fl->fl_owner);
230 /* set wait bit as appropriate, then make command as Ceph expects it*/
232 op = CEPH_MDS_OP_GETFILELOCK;
233 else if (IS_SETLKW(cmd))
236 spin_lock(&ci->i_ceph_lock);
237 if (ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) {
240 spin_unlock(&ci->i_ceph_lock);
242 if (op == CEPH_MDS_OP_SETFILELOCK && F_UNLCK == fl->fl_type)
243 posix_lock_file(file, fl, NULL);
247 if (F_RDLCK == fl->fl_type)
248 lock_cmd = CEPH_LOCK_SHARED;
249 else if (F_WRLCK == fl->fl_type)
250 lock_cmd = CEPH_LOCK_EXCL;
252 lock_cmd = CEPH_LOCK_UNLOCK;
254 err = ceph_lock_message(CEPH_LOCK_FCNTL, op, inode, lock_cmd, wait, fl);
256 if (op == CEPH_MDS_OP_SETFILELOCK) {
257 dout("mds locked, locking locally\n");
258 err = posix_lock_file(file, fl, NULL);
260 /* undo! This should only happen if
261 * the kernel detects local
263 ceph_lock_message(CEPH_LOCK_FCNTL, op, inode,
264 CEPH_LOCK_UNLOCK, 0, fl);
265 dout("got %d on posix_lock_file, undid lock\n",
273 int ceph_flock(struct file *file, int cmd, struct file_lock *fl)
275 struct inode *inode = file_inode(file);
276 struct ceph_inode_info *ci = ceph_inode(inode);
281 if (!(fl->fl_flags & FL_FLOCK))
283 /* No mandatory locks */
284 if (fl->fl_type & LOCK_MAND)
287 dout("ceph_flock, fl_file: %p\n", fl->fl_file);
289 spin_lock(&ci->i_ceph_lock);
290 if (ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) {
293 spin_unlock(&ci->i_ceph_lock);
295 if (F_UNLCK == fl->fl_type)
296 locks_lock_file_wait(file, fl);
303 if (F_RDLCK == fl->fl_type)
304 lock_cmd = CEPH_LOCK_SHARED;
305 else if (F_WRLCK == fl->fl_type)
306 lock_cmd = CEPH_LOCK_EXCL;
308 lock_cmd = CEPH_LOCK_UNLOCK;
310 err = ceph_lock_message(CEPH_LOCK_FLOCK, CEPH_MDS_OP_SETFILELOCK,
311 inode, lock_cmd, wait, fl);
313 err = locks_lock_file_wait(file, fl);
315 ceph_lock_message(CEPH_LOCK_FLOCK,
316 CEPH_MDS_OP_SETFILELOCK,
317 inode, CEPH_LOCK_UNLOCK, 0, fl);
318 dout("got %d on locks_lock_file_wait, undid lock\n", err);
325 * Fills in the passed counter variables, so you can prepare pagelist metadata
326 * before calling ceph_encode_locks.
328 void ceph_count_locks(struct inode *inode, int *fcntl_count, int *flock_count)
330 struct file_lock *lock;
331 struct file_lock_context *ctx;
336 ctx = inode->i_flctx;
338 spin_lock(&ctx->flc_lock);
339 list_for_each_entry(lock, &ctx->flc_posix, fl_list)
341 list_for_each_entry(lock, &ctx->flc_flock, fl_list)
343 spin_unlock(&ctx->flc_lock);
345 dout("counted %d flock locks and %d fcntl locks\n",
346 *flock_count, *fcntl_count);
350 * Given a pointer to a lock, convert it to a ceph filelock
352 static int lock_to_ceph_filelock(struct file_lock *lock,
353 struct ceph_filelock *cephlock)
356 cephlock->start = cpu_to_le64(lock->fl_start);
357 cephlock->length = cpu_to_le64(lock->fl_end - lock->fl_start + 1);
358 cephlock->client = cpu_to_le64(0);
359 cephlock->pid = cpu_to_le64((u64)lock->fl_pid);
360 cephlock->owner = cpu_to_le64(secure_addr(lock->fl_owner));
362 switch (lock->fl_type) {
364 cephlock->type = CEPH_LOCK_SHARED;
367 cephlock->type = CEPH_LOCK_EXCL;
370 cephlock->type = CEPH_LOCK_UNLOCK;
373 dout("Have unknown lock type %d\n", lock->fl_type);
381 * Encode the flock and fcntl locks for the given inode into the ceph_filelock
382 * array. Must be called with inode->i_lock already held.
383 * If we encounter more of a specific lock type than expected, return -ENOSPC.
385 int ceph_encode_locks_to_buffer(struct inode *inode,
386 struct ceph_filelock *flocks,
387 int num_fcntl_locks, int num_flock_locks)
389 struct file_lock *lock;
390 struct file_lock_context *ctx = inode->i_flctx;
396 dout("encoding %d flock and %d fcntl locks\n", num_flock_locks,
402 spin_lock(&ctx->flc_lock);
403 list_for_each_entry(lock, &ctx->flc_posix, fl_list) {
405 if (seen_fcntl > num_fcntl_locks) {
409 err = lock_to_ceph_filelock(lock, &flocks[l]);
414 list_for_each_entry(lock, &ctx->flc_flock, fl_list) {
416 if (seen_flock > num_flock_locks) {
420 err = lock_to_ceph_filelock(lock, &flocks[l]);
426 spin_unlock(&ctx->flc_lock);
431 * Copy the encoded flock and fcntl locks into the pagelist.
432 * Format is: #fcntl locks, sequential fcntl locks, #flock locks,
433 * sequential flock locks.
434 * Returns zero on success.
436 int ceph_locks_to_pagelist(struct ceph_filelock *flocks,
437 struct ceph_pagelist *pagelist,
438 int num_fcntl_locks, int num_flock_locks)
443 nlocks = cpu_to_le32(num_fcntl_locks);
444 err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
448 if (num_fcntl_locks > 0) {
449 err = ceph_pagelist_append(pagelist, flocks,
450 num_fcntl_locks * sizeof(*flocks));
455 nlocks = cpu_to_le32(num_flock_locks);
456 err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
460 if (num_flock_locks > 0) {
461 err = ceph_pagelist_append(pagelist, &flocks[num_fcntl_locks],
462 num_flock_locks * sizeof(*flocks));