Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0
0002 #include <linux/ceph/ceph_debug.h>
0003 
0004 #include <linux/file.h>
0005 #include <linux/namei.h>
0006 #include <linux/random.h>
0007 
0008 #include "super.h"
0009 #include "mds_client.h"
0010 #include <linux/ceph/pagelist.h>
0011 
0012 static u64 lock_secret;
0013 static int ceph_lock_wait_for_completion(struct ceph_mds_client *mdsc,
0014                                          struct ceph_mds_request *req);
0015 
0016 static inline u64 secure_addr(void *addr)
0017 {
0018     u64 v = lock_secret ^ (u64)(unsigned long)addr;
0019     /*
0020      * Set the most significant bit, so that MDS knows the 'owner'
0021      * is sufficient to identify the owner of lock. (old code uses
0022      * both 'owner' and 'pid')
0023      */
0024     v |= (1ULL << 63);
0025     return v;
0026 }
0027 
0028 void __init ceph_flock_init(void)
0029 {
0030     get_random_bytes(&lock_secret, sizeof(lock_secret));
0031 }
0032 
0033 static void ceph_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
0034 {
0035     struct ceph_file_info *fi = dst->fl_file->private_data;
0036     struct inode *inode = file_inode(dst->fl_file);
0037     atomic_inc(&ceph_inode(inode)->i_filelock_ref);
0038     atomic_inc(&fi->num_locks);
0039 }
0040 
0041 static void ceph_fl_release_lock(struct file_lock *fl)
0042 {
0043     struct ceph_file_info *fi = fl->fl_file->private_data;
0044     struct inode *inode = file_inode(fl->fl_file);
0045     struct ceph_inode_info *ci = ceph_inode(inode);
0046     atomic_dec(&fi->num_locks);
0047     if (atomic_dec_and_test(&ci->i_filelock_ref)) {
0048         /* clear error when all locks are released */
0049         spin_lock(&ci->i_ceph_lock);
0050         ci->i_ceph_flags &= ~CEPH_I_ERROR_FILELOCK;
0051         spin_unlock(&ci->i_ceph_lock);
0052     }
0053 }
0054 
0055 static const struct file_lock_operations ceph_fl_lock_ops = {
0056     .fl_copy_lock = ceph_fl_copy_lock,
0057     .fl_release_private = ceph_fl_release_lock,
0058 };
0059 
0060 /*
0061  * Implement fcntl and flock locking functions.
0062  */
0063 static int ceph_lock_message(u8 lock_type, u16 operation, struct inode *inode,
0064                  int cmd, u8 wait, struct file_lock *fl)
0065 {
0066     struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
0067     struct ceph_mds_request *req;
0068     int err;
0069     u64 length = 0;
0070     u64 owner;
0071 
0072     if (operation == CEPH_MDS_OP_SETFILELOCK) {
0073         /*
0074          * increasing i_filelock_ref closes race window between
0075          * handling request reply and adding file_lock struct to
0076          * inode. Otherwise, auth caps may get trimmed in the
0077          * window. Caller function will decrease the counter.
0078          */
0079         fl->fl_ops = &ceph_fl_lock_ops;
0080         fl->fl_ops->fl_copy_lock(fl, NULL);
0081     }
0082 
0083     if (operation != CEPH_MDS_OP_SETFILELOCK || cmd == CEPH_LOCK_UNLOCK)
0084         wait = 0;
0085 
0086     req = ceph_mdsc_create_request(mdsc, operation, USE_AUTH_MDS);
0087     if (IS_ERR(req))
0088         return PTR_ERR(req);
0089     req->r_inode = inode;
0090     ihold(inode);
0091     req->r_num_caps = 1;
0092 
0093     /* mds requires start and length rather than start and end */
0094     if (LLONG_MAX == fl->fl_end)
0095         length = 0;
0096     else
0097         length = fl->fl_end - fl->fl_start + 1;
0098 
0099     owner = secure_addr(fl->fl_owner);
0100 
0101     dout("ceph_lock_message: rule: %d, op: %d, owner: %llx, pid: %llu, "
0102          "start: %llu, length: %llu, wait: %d, type: %d\n", (int)lock_type,
0103          (int)operation, owner, (u64)fl->fl_pid, fl->fl_start, length,
0104          wait, fl->fl_type);
0105 
0106     req->r_args.filelock_change.rule = lock_type;
0107     req->r_args.filelock_change.type = cmd;
0108     req->r_args.filelock_change.owner = cpu_to_le64(owner);
0109     req->r_args.filelock_change.pid = cpu_to_le64((u64)fl->fl_pid);
0110     req->r_args.filelock_change.start = cpu_to_le64(fl->fl_start);
0111     req->r_args.filelock_change.length = cpu_to_le64(length);
0112     req->r_args.filelock_change.wait = wait;
0113 
0114     err = ceph_mdsc_submit_request(mdsc, inode, req);
0115     if (!err)
0116         err = ceph_mdsc_wait_request(mdsc, req, wait ?
0117                     ceph_lock_wait_for_completion : NULL);
0118     if (!err && operation == CEPH_MDS_OP_GETFILELOCK) {
0119         fl->fl_pid = -le64_to_cpu(req->r_reply_info.filelock_reply->pid);
0120         if (CEPH_LOCK_SHARED == req->r_reply_info.filelock_reply->type)
0121             fl->fl_type = F_RDLCK;
0122         else if (CEPH_LOCK_EXCL == req->r_reply_info.filelock_reply->type)
0123             fl->fl_type = F_WRLCK;
0124         else
0125             fl->fl_type = F_UNLCK;
0126 
0127         fl->fl_start = le64_to_cpu(req->r_reply_info.filelock_reply->start);
0128         length = le64_to_cpu(req->r_reply_info.filelock_reply->start) +
0129                          le64_to_cpu(req->r_reply_info.filelock_reply->length);
0130         if (length >= 1)
0131             fl->fl_end = length -1;
0132         else
0133             fl->fl_end = 0;
0134 
0135     }
0136     ceph_mdsc_put_request(req);
0137     dout("ceph_lock_message: rule: %d, op: %d, pid: %llu, start: %llu, "
0138          "length: %llu, wait: %d, type: %d, err code %d\n", (int)lock_type,
0139          (int)operation, (u64)fl->fl_pid, fl->fl_start,
0140          length, wait, fl->fl_type, err);
0141     return err;
0142 }
0143 
0144 static int ceph_lock_wait_for_completion(struct ceph_mds_client *mdsc,
0145                                          struct ceph_mds_request *req)
0146 {
0147     struct ceph_mds_request *intr_req;
0148     struct inode *inode = req->r_inode;
0149     int err, lock_type;
0150 
0151     BUG_ON(req->r_op != CEPH_MDS_OP_SETFILELOCK);
0152     if (req->r_args.filelock_change.rule == CEPH_LOCK_FCNTL)
0153         lock_type = CEPH_LOCK_FCNTL_INTR;
0154     else if (req->r_args.filelock_change.rule == CEPH_LOCK_FLOCK)
0155         lock_type = CEPH_LOCK_FLOCK_INTR;
0156     else
0157         BUG_ON(1);
0158     BUG_ON(req->r_args.filelock_change.type == CEPH_LOCK_UNLOCK);
0159 
0160     err = wait_for_completion_interruptible(&req->r_completion);
0161     if (!err)
0162         return 0;
0163 
0164     dout("ceph_lock_wait_for_completion: request %llu was interrupted\n",
0165          req->r_tid);
0166 
0167     mutex_lock(&mdsc->mutex);
0168     if (test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
0169         err = 0;
0170     } else {
0171         /*
0172          * ensure we aren't running concurrently with
0173          * ceph_fill_trace or ceph_readdir_prepopulate, which
0174          * rely on locks (dir mutex) held by our caller.
0175          */
0176         mutex_lock(&req->r_fill_mutex);
0177         req->r_err = err;
0178         set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
0179         mutex_unlock(&req->r_fill_mutex);
0180 
0181         if (!req->r_session) {
0182             // haven't sent the request
0183             err = 0;
0184         }
0185     }
0186     mutex_unlock(&mdsc->mutex);
0187     if (!err)
0188         return 0;
0189 
0190     intr_req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETFILELOCK,
0191                         USE_AUTH_MDS);
0192     if (IS_ERR(intr_req))
0193         return PTR_ERR(intr_req);
0194 
0195     intr_req->r_inode = inode;
0196     ihold(inode);
0197     intr_req->r_num_caps = 1;
0198 
0199     intr_req->r_args.filelock_change = req->r_args.filelock_change;
0200     intr_req->r_args.filelock_change.rule = lock_type;
0201     intr_req->r_args.filelock_change.type = CEPH_LOCK_UNLOCK;
0202 
0203     err = ceph_mdsc_do_request(mdsc, inode, intr_req);
0204     ceph_mdsc_put_request(intr_req);
0205 
0206     if (err && err != -ERESTARTSYS)
0207         return err;
0208 
0209     wait_for_completion_killable(&req->r_safe_completion);
0210     return 0;
0211 }
0212 
0213 static int try_unlock_file(struct file *file, struct file_lock *fl)
0214 {
0215     int err;
0216     unsigned int orig_flags = fl->fl_flags;
0217     fl->fl_flags |= FL_EXISTS;
0218     err = locks_lock_file_wait(file, fl);
0219     fl->fl_flags = orig_flags;
0220     if (err == -ENOENT) {
0221         if (!(orig_flags & FL_EXISTS))
0222             err = 0;
0223         return err;
0224     }
0225     return 1;
0226 }
0227 
0228 /*
0229  * Attempt to set an fcntl lock.
0230  * For now, this just goes away to the server. Later it may be more awesome.
0231  */
0232 int ceph_lock(struct file *file, int cmd, struct file_lock *fl)
0233 {
0234     struct inode *inode = file_inode(file);
0235     struct ceph_inode_info *ci = ceph_inode(inode);
0236     int err = 0;
0237     u16 op = CEPH_MDS_OP_SETFILELOCK;
0238     u8 wait = 0;
0239     u8 lock_cmd;
0240 
0241     if (!(fl->fl_flags & FL_POSIX))
0242         return -ENOLCK;
0243 
0244     if (ceph_inode_is_shutdown(inode))
0245         return -ESTALE;
0246 
0247     dout("ceph_lock, fl_owner: %p\n", fl->fl_owner);
0248 
0249     /* set wait bit as appropriate, then make command as Ceph expects it*/
0250     if (IS_GETLK(cmd))
0251         op = CEPH_MDS_OP_GETFILELOCK;
0252     else if (IS_SETLKW(cmd))
0253         wait = 1;
0254 
0255     spin_lock(&ci->i_ceph_lock);
0256     if (ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) {
0257         err = -EIO;
0258     }
0259     spin_unlock(&ci->i_ceph_lock);
0260     if (err < 0) {
0261         if (op == CEPH_MDS_OP_SETFILELOCK && F_UNLCK == fl->fl_type)
0262             posix_lock_file(file, fl, NULL);
0263         return err;
0264     }
0265 
0266     if (F_RDLCK == fl->fl_type)
0267         lock_cmd = CEPH_LOCK_SHARED;
0268     else if (F_WRLCK == fl->fl_type)
0269         lock_cmd = CEPH_LOCK_EXCL;
0270     else
0271         lock_cmd = CEPH_LOCK_UNLOCK;
0272 
0273     if (op == CEPH_MDS_OP_SETFILELOCK && F_UNLCK == fl->fl_type) {
0274         err = try_unlock_file(file, fl);
0275         if (err <= 0)
0276             return err;
0277     }
0278 
0279     err = ceph_lock_message(CEPH_LOCK_FCNTL, op, inode, lock_cmd, wait, fl);
0280     if (!err) {
0281         if (op == CEPH_MDS_OP_SETFILELOCK && F_UNLCK != fl->fl_type) {
0282             dout("mds locked, locking locally\n");
0283             err = posix_lock_file(file, fl, NULL);
0284             if (err) {
0285                 /* undo! This should only happen if
0286                  * the kernel detects local
0287                  * deadlock. */
0288                 ceph_lock_message(CEPH_LOCK_FCNTL, op, inode,
0289                           CEPH_LOCK_UNLOCK, 0, fl);
0290                 dout("got %d on posix_lock_file, undid lock\n",
0291                      err);
0292             }
0293         }
0294     }
0295     return err;
0296 }
0297 
0298 int ceph_flock(struct file *file, int cmd, struct file_lock *fl)
0299 {
0300     struct inode *inode = file_inode(file);
0301     struct ceph_inode_info *ci = ceph_inode(inode);
0302     int err = 0;
0303     u8 wait = 0;
0304     u8 lock_cmd;
0305 
0306     if (!(fl->fl_flags & FL_FLOCK))
0307         return -ENOLCK;
0308 
0309     if (ceph_inode_is_shutdown(inode))
0310         return -ESTALE;
0311 
0312     dout("ceph_flock, fl_file: %p\n", fl->fl_file);
0313 
0314     spin_lock(&ci->i_ceph_lock);
0315     if (ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) {
0316         err = -EIO;
0317     }
0318     spin_unlock(&ci->i_ceph_lock);
0319     if (err < 0) {
0320         if (F_UNLCK == fl->fl_type)
0321             locks_lock_file_wait(file, fl);
0322         return err;
0323     }
0324 
0325     if (IS_SETLKW(cmd))
0326         wait = 1;
0327 
0328     if (F_RDLCK == fl->fl_type)
0329         lock_cmd = CEPH_LOCK_SHARED;
0330     else if (F_WRLCK == fl->fl_type)
0331         lock_cmd = CEPH_LOCK_EXCL;
0332     else
0333         lock_cmd = CEPH_LOCK_UNLOCK;
0334 
0335     if (F_UNLCK == fl->fl_type) {
0336         err = try_unlock_file(file, fl);
0337         if (err <= 0)
0338             return err;
0339     }
0340 
0341     err = ceph_lock_message(CEPH_LOCK_FLOCK, CEPH_MDS_OP_SETFILELOCK,
0342                 inode, lock_cmd, wait, fl);
0343     if (!err && F_UNLCK != fl->fl_type) {
0344         err = locks_lock_file_wait(file, fl);
0345         if (err) {
0346             ceph_lock_message(CEPH_LOCK_FLOCK,
0347                       CEPH_MDS_OP_SETFILELOCK,
0348                       inode, CEPH_LOCK_UNLOCK, 0, fl);
0349             dout("got %d on locks_lock_file_wait, undid lock\n", err);
0350         }
0351     }
0352     return err;
0353 }
0354 
0355 /*
0356  * Fills in the passed counter variables, so you can prepare pagelist metadata
0357  * before calling ceph_encode_locks.
0358  */
0359 void ceph_count_locks(struct inode *inode, int *fcntl_count, int *flock_count)
0360 {
0361     struct file_lock *lock;
0362     struct file_lock_context *ctx;
0363 
0364     *fcntl_count = 0;
0365     *flock_count = 0;
0366 
0367     ctx = inode->i_flctx;
0368     if (ctx) {
0369         spin_lock(&ctx->flc_lock);
0370         list_for_each_entry(lock, &ctx->flc_posix, fl_list)
0371             ++(*fcntl_count);
0372         list_for_each_entry(lock, &ctx->flc_flock, fl_list)
0373             ++(*flock_count);
0374         spin_unlock(&ctx->flc_lock);
0375     }
0376     dout("counted %d flock locks and %d fcntl locks\n",
0377          *flock_count, *fcntl_count);
0378 }
0379 
0380 /*
0381  * Given a pointer to a lock, convert it to a ceph filelock
0382  */
0383 static int lock_to_ceph_filelock(struct file_lock *lock,
0384                  struct ceph_filelock *cephlock)
0385 {
0386     int err = 0;
0387     cephlock->start = cpu_to_le64(lock->fl_start);
0388     cephlock->length = cpu_to_le64(lock->fl_end - lock->fl_start + 1);
0389     cephlock->client = cpu_to_le64(0);
0390     cephlock->pid = cpu_to_le64((u64)lock->fl_pid);
0391     cephlock->owner = cpu_to_le64(secure_addr(lock->fl_owner));
0392 
0393     switch (lock->fl_type) {
0394     case F_RDLCK:
0395         cephlock->type = CEPH_LOCK_SHARED;
0396         break;
0397     case F_WRLCK:
0398         cephlock->type = CEPH_LOCK_EXCL;
0399         break;
0400     case F_UNLCK:
0401         cephlock->type = CEPH_LOCK_UNLOCK;
0402         break;
0403     default:
0404         dout("Have unknown lock type %d\n", lock->fl_type);
0405         err = -EINVAL;
0406     }
0407 
0408     return err;
0409 }
0410 
0411 /*
0412  * Encode the flock and fcntl locks for the given inode into the ceph_filelock
0413  * array. Must be called with inode->i_lock already held.
0414  * If we encounter more of a specific lock type than expected, return -ENOSPC.
0415  */
0416 int ceph_encode_locks_to_buffer(struct inode *inode,
0417                 struct ceph_filelock *flocks,
0418                 int num_fcntl_locks, int num_flock_locks)
0419 {
0420     struct file_lock *lock;
0421     struct file_lock_context *ctx = inode->i_flctx;
0422     int err = 0;
0423     int seen_fcntl = 0;
0424     int seen_flock = 0;
0425     int l = 0;
0426 
0427     dout("encoding %d flock and %d fcntl locks\n", num_flock_locks,
0428          num_fcntl_locks);
0429 
0430     if (!ctx)
0431         return 0;
0432 
0433     spin_lock(&ctx->flc_lock);
0434     list_for_each_entry(lock, &ctx->flc_posix, fl_list) {
0435         ++seen_fcntl;
0436         if (seen_fcntl > num_fcntl_locks) {
0437             err = -ENOSPC;
0438             goto fail;
0439         }
0440         err = lock_to_ceph_filelock(lock, &flocks[l]);
0441         if (err)
0442             goto fail;
0443         ++l;
0444     }
0445     list_for_each_entry(lock, &ctx->flc_flock, fl_list) {
0446         ++seen_flock;
0447         if (seen_flock > num_flock_locks) {
0448             err = -ENOSPC;
0449             goto fail;
0450         }
0451         err = lock_to_ceph_filelock(lock, &flocks[l]);
0452         if (err)
0453             goto fail;
0454         ++l;
0455     }
0456 fail:
0457     spin_unlock(&ctx->flc_lock);
0458     return err;
0459 }
0460 
0461 /*
0462  * Copy the encoded flock and fcntl locks into the pagelist.
0463  * Format is: #fcntl locks, sequential fcntl locks, #flock locks,
0464  * sequential flock locks.
0465  * Returns zero on success.
0466  */
0467 int ceph_locks_to_pagelist(struct ceph_filelock *flocks,
0468                struct ceph_pagelist *pagelist,
0469                int num_fcntl_locks, int num_flock_locks)
0470 {
0471     int err = 0;
0472     __le32 nlocks;
0473 
0474     nlocks = cpu_to_le32(num_fcntl_locks);
0475     err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
0476     if (err)
0477         goto out_fail;
0478 
0479     if (num_fcntl_locks > 0) {
0480         err = ceph_pagelist_append(pagelist, flocks,
0481                        num_fcntl_locks * sizeof(*flocks));
0482         if (err)
0483             goto out_fail;
0484     }
0485 
0486     nlocks = cpu_to_le32(num_flock_locks);
0487     err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
0488     if (err)
0489         goto out_fail;
0490 
0491     if (num_flock_locks > 0) {
0492         err = ceph_pagelist_append(pagelist, &flocks[num_fcntl_locks],
0493                        num_flock_locks * sizeof(*flocks));
0494     }
0495 out_fail:
0496     return err;
0497 }
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!