Messages in this thread Patch in this message | | | Date | Thu, 17 Jun 1999 17:13:12 +0200 | From | Jan Kara <> | Subject | [FIX] Quotas |
| |
Hello.
Here are my fixes for quotas for 2.2.10. Changes against 2.2.9 are:
* Added locking when walking through inode lists (inode_lock). * Check for bogus uid/gid in quotactl -- ie. id > 65535 * Fixed check for group in quota
Fixes for 2.2.9 were:
* Added check to dqget() so it now can't return invalidated quota (could happen when quota was invalidated while we were sleeping). * Fixed races in dquot_transfer -- quota now locked before testing and unlocked after updating so it can't change in the mean time. * Same for dquot_(free/alloc)_(block/inode). * Fixed race in DQUOT_TRANSFER - we could move quota from user A to user B, then find out that notify_change() can't be done so we want to move quota back but we can't -- ie. quota is reached and we leave quota to user B. Might be even exploitable when we think about groups. * invalidate_dquot() writes modified quotas -- when invalidate is in process somebody might hold quota too (some operation like get_quota) so it needn't to be written by dqput. * sync_dquot did only one pass (on the second one the pointer wasn't initialized). Fixed. * sync_dquot could call write_dquot on invalidated dquot. Fixed. * When there was no memory for new dquot but we hadn't reached MAX_DQUOTS, we were trying to allocate some memory in tight loop forever... * grow_dquots() now initializes wait_queue correctly * add_dquot_ref() is restarting after each blocking operation (filp list might change) * reset_dquot_ptrs() was going just through filps so we could easily forget some quota in i-cache Now reset_dquot_ptrs() is going through inode lists. * dqput() didn't check whether quota isn't invalidated after blocking which could result in NULL dereference. * write_dquot() might need to update written quota (when we were allocating next block for our file) which resulted in deadlock (I can't understand how there could exist 2 people seeing it :-)). * (read/write)_dquot didn't check, whether quota wasn't invalidated while they were sleeping. Ooops. * There could be races between concurrent quota_ons and quota_offs (quite unprobable but theoretically possible). Added new dqoff_sem semaphore serializing quota_ons and quota_offs. * Some other minor cleanup...
Patches can be downloaded at ftp://atrey.karlin.mff.cuni.cz/pub/local/jack/. quota-fix-2.2.10-1.diff.gz is patch against clean 2.2.10, quota-fix-2.2.10-1a.diff is against kernel with my previous patches. Patch against clean kernel is also attached.
Linus, please, apply my patch (or at least mail me, what you dislike... :)).
Honza.
--- linux/fs/dquot.c Wed Jun 16 21:54:33 1999 +++ linux/fs/dquot.c Wed Jun 16 22:39:30 1999 @@ -21,6 +21,17 @@ * Revised list management to avoid races * -- Bill Hawes, <whawes@star.net>, 9/98 * + * Fixed races in dquot_transfer(), dqget() and dquot_alloc_...(). + * As the consequence the locking was moved from dquot_decr_...(), + * dquot_incr_...() to calling functions. + * invalidate_dquots() now writes modified dquots. + * Serialized quota_off() and quota_on() for mount point. + * Fixed a few bugs in grow_dquots. Fixed deadlock in write_dquot(). + * reset_dquot_ptrs() now traverse through inodes not filps. + * add_dquot_ref() restarts after blocking + * Added check for bogus uid and fixed check for group in quotactl. + * Jan Kara, <jack@atrey.karlin.mff.cuni.cz>, 4-6/99 + * * (C) Copyright 1994 - 1997 Marco van Wieringen */ @@ -48,6 +59,10 @@ int nr_dquots = 0, nr_free_dquots = 0; int max_dquots = NR_DQUOTS; +/* We need this list for invalidating dquots... */ +extern struct list_head inode_in_use; +extern spinlock_t inode_lock; + static char quotamessage[MAX_QUOTA_MESSAGE]; static char *quotatypes[] = INITQFNAMES; @@ -166,7 +181,8 @@ { /* sanity check */ if (list_empty(&dquot->dq_free)) { - printk("remove_free_dquot: dquot not on free list??\n"); + printk("remove_free_dquot: dquot not on the free list??\n"); + return; /* J.K. Just don't do anything */ } list_del(&dquot->dq_free); INIT_LIST_HEAD(&dquot->dq_free); @@ -229,13 +245,25 @@ static void write_dquot(struct dquot *dquot) { short type = dquot->dq_type; - struct file *filp = dquot->dq_mnt->mnt_dquot.files[type]; + struct file *filp; mm_segment_t fs; loff_t offset; ssize_t ret; + struct dqblk data; + struct semaphore *sem = &dquot->dq_mnt->mnt_dquot.dqio_sem; - lock_dquot(dquot); - down(&dquot->dq_mnt->mnt_dquot.semaphore); + /* + * We copy our data to preserve consistency when dquot is not locked. + * We can't lock it because it can cause deadlocks - think about + * growing the quota file... + */ + memcpy(&data, &dquot->dq_dqb, sizeof(struct dqblk)); + down(sem); + if (!dquot->dq_mnt) { /* Invalidated quota? */ + up(sem); + return; + } + filp = dquot->dq_mnt->mnt_dquot.files[type]; offset = dqoff(dquot->dq_id); fs = get_fs(); set_fs(KERNEL_DS); @@ -247,46 +275,51 @@ dquot->dq_flags &= ~DQ_MOD; ret = 0; if (filp) - ret = filp->f_op->write(filp, (char *)&dquot->dq_dqb, + ret = filp->f_op->write(filp, (char *)&data, sizeof(struct dqblk), &offset); if (ret != sizeof(struct dqblk)) printk(KERN_WARNING "VFS: dquota write failed on dev %s\n", kdevname(dquot->dq_dev)); - up(&dquot->dq_mnt->mnt_dquot.semaphore); + /* + * Be sure that anybody didn't invalidated the dquot in the mean time... + * Nasty but I don't see other choice when we don't want to lock dquot. + */ + up(sem); set_fs(fs); - unlock_dquot(dquot); dqstats.writes++; } static void read_dquot(struct dquot *dquot) { - short type; + short type = dquot->dq_type; struct file *filp; mm_segment_t fs; loff_t offset; - type = dquot->dq_type; filp = dquot->dq_mnt->mnt_dquot.files[type]; - if (filp == (struct file *)NULL) return; lock_dquot(dquot); - down(&dquot->dq_mnt->mnt_dquot.semaphore); + if (!dquot->dq_mnt) /* Invalidated quota? */ + goto out_lock; + /* Now we are sure filp is valid - the dquot isn't invalidated */ + down(&dquot->dq_mnt->mnt_dquot.dqio_sem); offset = dqoff(dquot->dq_id); fs = get_fs(); set_fs(KERNEL_DS); filp->f_op->read(filp, (char *)&dquot->dq_dqb, sizeof(struct dqblk), &offset); - up(&dquot->dq_mnt->mnt_dquot.semaphore); + up(&dquot->dq_mnt->mnt_dquot.dqio_sem); set_fs(fs); if (dquot->dq_bhardlimit == 0 && dquot->dq_bsoftlimit == 0 && dquot->dq_ihardlimit == 0 && dquot->dq_isoftlimit == 0) dquot->dq_flags |= DQ_FAKE; - unlock_dquot(dquot); dqstats.reads++; +out_lock: + unlock_dquot(dquot); } /* @@ -306,10 +339,11 @@ void invalidate_dquots(kdev_t dev, short type) { - struct dquot *dquot, *next = inuse_list; + struct dquot *dquot, *next; int need_restart; restart: + next = inuse_list; /* Here it is better. Otherwise the restart doesn't have any sense ;-) */ need_restart = 0; while ((dquot = next) != NULL) { next = dquot->dq_next; @@ -317,6 +351,8 @@ continue; if (dquot->dq_type != type) continue; + if (!dquot->dq_mnt) /* Already invalidated entry? */ + continue; if (dquot->dq_flags & DQ_LOCKED) { __wait_on_dquot(dquot); @@ -329,6 +365,18 @@ continue; if (dquot->dq_type != type) continue; + if (!dquot->dq_mnt) + continue; + } + /* + * Because inodes needn't to be the only holders of dquot + * the quota needn't to be written to disk. So we write it + * ourselves before discarding the data just for sure... + */ + if (dquot->dq_flags & DQ_MOD && dquot->dq_mnt) + { + write_dquot(dquot); + need_restart = 1; /* We slept on IO */ } clear_dquot(dquot); } @@ -342,10 +390,11 @@ int sync_dquots(kdev_t dev, short type) { - struct dquot *dquot, *next = inuse_list; + struct dquot *dquot, *next; int need_restart; restart: + next = inuse_list; need_restart = 0; while ((dquot = next) != NULL) { next = dquot->dq_next; @@ -353,6 +402,8 @@ continue; if (type != -1 && dquot->dq_type != type) continue; + if (!dquot->dq_mnt) /* Invalidated? */ + continue; if (!(dquot->dq_flags & (DQ_LOCKED | DQ_MOD))) continue; @@ -390,10 +441,11 @@ * checking and doesn't need to be written. It's just an empty * dquot that is put back on to the freelist. */ - if (dquot->dq_mnt != (struct vfsmount *)NULL) { + if (dquot->dq_mnt) dqstats.drops++; we_slept: - wait_on_dquot(dquot); + wait_on_dquot(dquot); + if (dquot->dq_mnt) { if (dquot->dq_count > 1) { dquot->dq_count--; return; @@ -407,8 +459,16 @@ /* sanity check */ if (!list_empty(&dquot->dq_free)) { printk("dqput: dquot already on free list??\n"); + dquot->dq_count--; /* J.K. Just decrementing use count seems safer... */ + return; } if (--dquot->dq_count == 0) { + /* Sanity check. Locked quota without owner isn't good idea... */ + if (dquot->dq_flags & DQ_LOCKED) { + printk(KERN_ERR "VFS: Locked quota to be put on the free list.\n"); + dquot->dq_flags &= ~DQ_LOCKED; + } + dquot->dq_flags &= ~DQ_MOD; /* Modified flag has no sense on free list */ /* Place at end of LRU free queue */ put_dquot_last(dquot); wake_up(&dquot_wait); @@ -417,23 +477,25 @@ return; } -static void grow_dquots(void) +static int grow_dquots(void) { struct dquot *dquot; - int cnt = 32; + int cnt = 0; - while (cnt > 0) { + while (cnt < 32) { dquot = kmem_cache_alloc(dquot_cachep, SLAB_KERNEL); if(!dquot) - return; + return cnt; nr_dquots++; memset((caddr_t)dquot, 0, sizeof(struct dquot)); + init_waitqueue(&dquot->dq_wait); /* all dquots go on the inuse_list */ put_inuse(dquot); put_dquot_head(dquot); - cnt--; + cnt++; } + return cnt; } static struct dquot *find_best_candidate_weighted(void) @@ -445,6 +507,7 @@ while ((tmp = tmp->next) != &free_dquots && --limit) { dquot = list_entry(tmp, struct dquot, dq_free); + /* This should never happen... */ if (dquot->dq_flags & (DQ_LOCKED | DQ_MOD)) continue; myscore = dquot->dq_referenced; @@ -480,20 +543,9 @@ if (!dquot) goto pressure; got_it: - if (dquot->dq_flags & (DQ_LOCKED | DQ_MOD)) { - wait_on_dquot(dquot); - if (dquot->dq_flags & DQ_MOD) - { - if(dquot->dq_mnt != (struct vfsmount *)NULL) - write_dquot(dquot); - } - /* - * The dquot may be back in use now, so we - * must recheck the free list. - */ - goto repeat; - } - /* sanity check ... */ + /* Sanity checks */ + if (dquot->dq_flags & DQ_LOCKED) + printk(KERN_ERR "VFS: Locked dquot on the free list\n"); if (dquot->dq_count != 0) printk(KERN_ERR "VFS: free dquot count=%d\n", dquot->dq_count); @@ -504,10 +556,9 @@ return dquot; pressure: - if (nr_dquots < max_dquots) { - grow_dquots(); - goto repeat; - } + if (nr_dquots < max_dquots) + if (grow_dquots()) + goto repeat; dquot = find_best_candidate_weighted(); if (dquot) @@ -534,7 +585,7 @@ struct dquot *dquot, *empty = NULL; struct vfsmount *vfsmnt; - if ((vfsmnt = lookup_vfsmnt(dev)) == (struct vfsmount *)NULL || is_enabled(vfsmnt, type) == 0) + if ((vfsmnt = lookup_vfsmnt(dev)) == (struct vfsmount *)NULL || !is_enabled(vfsmnt, type)) return(NODQUOT); we_slept: @@ -567,12 +618,46 @@ while (dquot_updating[hashent]) sleep_on(&update_wait); + if (!dquot->dq_mnt) { /* Has somebody invalidated entry under us? */ + /* + * Do it as if the quota was invalidated before we started + */ + dqput(dquot); + return NODQUOT; + } dquot->dq_referenced++; dqstats.lookups++; return dquot; } +static struct dquot *dqduplicate(struct dquot *dquot) +{ + if (dquot == NODQUOT || !dquot->dq_mnt) + return NODQUOT; + dquot->dq_count++; + wait_on_dquot(dquot); + if (!dquot->dq_mnt) { + dquot->dq_count--; + return NODQUOT; + } + return dquot; +} + +static int dqinit_needed(struct inode *inode, short type) +{ + int cnt; + + if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))) + return 0; + if (type != -1) + return inode->i_dquot[type] == NODQUOT; + for (cnt = 0; cnt < MAXQUOTAS; cnt++) + if (inode->i_dquot[cnt] == NODQUOT) + return 1; + return 0; +} + static void add_dquot_ref(kdev_t dev, short type) { struct super_block *sb = get_super(dev); @@ -582,6 +667,7 @@ if (!sb || !sb->dq_op) return; /* nothing to do */ +restart: for (filp = inuse_filps; filp; filp = filp->f_next) { if (!filp->f_dentry) continue; @@ -590,79 +676,97 @@ inode = filp->f_dentry->d_inode; if (!inode) continue; - /* N.B. race problem -- filp could become unused */ - if (filp->f_mode & FMODE_WRITE) { + /* Didn't we already initialized this inode? */ + if (filp->f_mode & FMODE_WRITE && dqinit_needed(inode, type)) { sb->dq_op->initialize(inode, type); inode->i_flags |= S_QUOTA; + /* as we may have blocked we had better restart */ + goto restart; } } } +static int reset_inode_dquot_ptrs(struct inode *inode, short type) +{ + struct dquot *dquot = inode->i_dquot[type]; + int cnt; + + inode->i_dquot[type] = NODQUOT; + /* any other quota in use? */ + for (cnt = 0; cnt < MAXQUOTAS; cnt++) { + if (inode->i_dquot[cnt] != NODQUOT) + goto put_it; + } + inode->i_flags &= ~S_QUOTA; +put_it: + if (dquot != NODQUOT) { + spin_unlock(&inode_lock); /* We may block so drop the lock... */ + dqput(dquot); + spin_lock(&inode_lock); /* And capture lock again */ + /* we may have blocked ... */ + return 1; + } + return 0; +} + static void reset_dquot_ptrs(kdev_t dev, short type) { struct super_block *sb = get_super(dev); - struct file *filp; struct inode *inode; - struct dquot *dquot; - int cnt; + struct list_head *act_head; + int need_list = 3; if (!sb || !sb->dq_op) return; /* nothing to do */ -restart: - /* free any quota for unused dentries */ - shrink_dcache_sb(sb); + /* We have to be protected against other CPUs */ + spin_lock(&inode_lock); - for (filp = inuse_filps; filp; filp = filp->f_next) { - if (!filp->f_dentry) - continue; - if (filp->f_dentry->d_sb != sb) - continue; - inode = filp->f_dentry->d_inode; - if (!inode) - continue; - /* - * Note: we restart after each blocking operation, - * as the inuse_filps list may have changed. - */ - if (IS_QUOTAINIT(inode)) { - dquot = inode->i_dquot[type]; - inode->i_dquot[type] = NODQUOT; - /* any other quota in use? */ - for (cnt = 0; cnt < MAXQUOTAS; cnt++) { - if (inode->i_dquot[cnt] != NODQUOT) - goto put_it; + do { + if (need_list & 1) { + need_list &= ~1; + restart_in_use: + for (act_head = inode_in_use.next; act_head != &inode_in_use; act_head = act_head->next) { + inode = list_entry(act_head, struct inode, i_list); + if (inode->i_sb != sb || !IS_QUOTAINIT(inode)) + continue; + if (reset_inode_dquot_ptrs(inode, type)) { + need_list |= 2; + goto restart_in_use; + } } - inode->i_flags &= ~S_QUOTA; - put_it: - if (dquot != NODQUOT) { - dqput(dquot); - /* we may have blocked ... */ - goto restart; + } + if (need_list & 2) { + need_list &= ~2; + restart_dirty: + for (act_head = sb->s_dirty.next; act_head != &sb->s_dirty; act_head = act_head->next) { + inode = list_entry(act_head, struct inode, i_list); + if (IS_QUOTAINIT(inode) && reset_inode_dquot_ptrs(inode, type)) { + need_list |= 1; + goto restart_dirty; + } } } } + while (need_list); + + spin_unlock(&inode_lock); } static inline void dquot_incr_inodes(struct dquot *dquot, unsigned long number) { - lock_dquot(dquot); dquot->dq_curinodes += number; dquot->dq_flags |= DQ_MOD; - unlock_dquot(dquot); } static inline void dquot_incr_blocks(struct dquot *dquot, unsigned long number) { - lock_dquot(dquot); dquot->dq_curblocks += number; dquot->dq_flags |= DQ_MOD; - unlock_dquot(dquot); } static inline void dquot_decr_inodes(struct dquot *dquot, unsigned long number) { - lock_dquot(dquot); if (dquot->dq_curinodes > number) dquot->dq_curinodes -= number; else @@ -671,12 +775,10 @@ dquot->dq_itime = (time_t) 0; dquot->dq_flags &= ~DQ_INODES; dquot->dq_flags |= DQ_MOD; - unlock_dquot(dquot); } static inline void dquot_decr_blocks(struct dquot *dquot, unsigned long number) { - lock_dquot(dquot); if (dquot->dq_curblocks > number) dquot->dq_curblocks -= number; else @@ -685,7 +787,6 @@ dquot->dq_btime = (time_t) 0; dquot->dq_flags &= ~DQ_BLKS; dquot->dq_flags |= DQ_MOD; - unlock_dquot(dquot); } static inline char need_print_warning(short type, uid_t initiator, struct dquot *dquot) @@ -704,7 +805,7 @@ return(initiator == 0 && dquot->dq_mnt->mnt_dquot.rsquash[dquot->dq_type] == 0); } -static int check_idq(struct dquot *dquot, short type, u_long short inodes, uid_t initiator, +static int check_idq(struct dquot *dquot, short type, u_long inodes, uid_t initiator, struct tty_struct *tty) { if (inodes <= 0 || dquot->dq_flags & DQ_FAKE) @@ -869,9 +970,11 @@ if (dquot == NODQUOT) goto out; + lock_dquot(dquot); /* We must protect against invalidating the quota */ error = -EFAULT; if (dqblk && !copy_to_user(dqblk, &dquot->dq_dqb, sizeof(struct dqblk))) error = 0; + unlock_dquot(dquot); dqput(dquot); out: return error; @@ -963,6 +1066,8 @@ break; } dquot = dqget(inode->i_dev, id, cnt); + if (dquot == NODQUOT) + continue; if (inode->i_dquot[cnt] != NODQUOT) { dqput(dquot); continue; @@ -1000,23 +1105,36 @@ int dquot_alloc_block(const struct inode *inode, unsigned long number, uid_t initiator, char warn) { - unsigned short cnt; + int cnt; struct tty_struct *tty = current->tty; + struct dquot *dquot[MAXQUOTAS]; for (cnt = 0; cnt < MAXQUOTAS; cnt++) { - if (inode->i_dquot[cnt] == NODQUOT) + dquot[cnt] = dqduplicate(inode->i_dquot[cnt]); + if (dquot[cnt] == NODQUOT) continue; - if (check_bdq(inode->i_dquot[cnt], cnt, number, initiator, tty, warn)) - return(NO_QUOTA); + lock_dquot(dquot[cnt]); + if (check_bdq(dquot[cnt], cnt, number, initiator, tty, warn)) + goto put_all; } for (cnt = 0; cnt < MAXQUOTAS; cnt++) { - if (inode->i_dquot[cnt] == NODQUOT) + if (dquot[cnt] == NODQUOT) continue; - dquot_incr_blocks(inode->i_dquot[cnt], number); + dquot_incr_blocks(dquot[cnt], number); + unlock_dquot(dquot[cnt]); + dqput(dquot[cnt]); } - return(QUOTA_OK); + return QUOTA_OK; +put_all: + for (; cnt >= 0; cnt--) { + if (dquot[cnt] == NODQUOT) + continue; + unlock_dquot(dquot[cnt]); + dqput(dquot[cnt]); + } + return NO_QUOTA; } /* @@ -1024,23 +1142,36 @@ */ int dquot_alloc_inode(const struct inode *inode, unsigned long number, uid_t initiator) { - unsigned short cnt; + int cnt; struct tty_struct *tty = current->tty; + struct dquot *dquot[MAXQUOTAS]; for (cnt = 0; cnt < MAXQUOTAS; cnt++) { - if (inode->i_dquot[cnt] == NODQUOT) + dquot[cnt] = dqduplicate(inode -> i_dquot[cnt]); + if (dquot[cnt] == NODQUOT) continue; - if (check_idq(inode->i_dquot[cnt], cnt, number, initiator, tty)) - return(NO_QUOTA); + lock_dquot(dquot[cnt]); + if (check_idq(dquot[cnt], cnt, number, initiator, tty)) + goto put_all; } for (cnt = 0; cnt < MAXQUOTAS; cnt++) { - if (inode->i_dquot[cnt] == NODQUOT) + if (dquot[cnt] == NODQUOT) continue; - dquot_incr_inodes(inode->i_dquot[cnt], number); + dquot_incr_inodes(dquot[cnt], number); + unlock_dquot(dquot[cnt]); + dqput(dquot[cnt]); } - return(QUOTA_OK); + return QUOTA_OK; +put_all: + for (; cnt >= 0; cnt--) { + if (dquot[cnt] == NODQUOT) + continue; + unlock_dquot(dquot[cnt]); + dqput(dquot[cnt]); + } + return NO_QUOTA; } /* @@ -1049,11 +1180,14 @@ void dquot_free_block(const struct inode *inode, unsigned long number) { unsigned short cnt; + struct dquot *dquot; for (cnt = 0; cnt < MAXQUOTAS; cnt++) { - if (inode->i_dquot[cnt] == NODQUOT) + dquot = inode->i_dquot[cnt]; + if (dquot == NODQUOT) continue; - dquot_decr_blocks(inode->i_dquot[cnt], number); + wait_on_dquot(dquot); + dquot_decr_blocks(dquot, number); } } @@ -1063,11 +1197,14 @@ void dquot_free_inode(const struct inode *inode, unsigned long number) { unsigned short cnt; + struct dquot *dquot; for (cnt = 0; cnt < MAXQUOTAS; cnt++) { - if (inode->i_dquot[cnt] == NODQUOT) + dquot = inode->i_dquot[cnt]; + if (dquot == NODQUOT) continue; - dquot_decr_inodes(inode->i_dquot[cnt], number); + wait_on_dquot(dquot); + dquot_decr_inodes(dquot, number); } } @@ -1076,14 +1213,18 @@ * * Note: this is a blocking operation. */ -int dquot_transfer(struct inode *inode, struct iattr *iattr, char direction, uid_t initiator) +int dquot_transfer(struct dentry *dentry, struct iattr *iattr, uid_t initiator) { + struct inode *inode = dentry -> d_inode; unsigned long blocks; struct dquot *transfer_from[MAXQUOTAS]; struct dquot *transfer_to[MAXQUOTAS]; struct tty_struct *tty = current->tty; short cnt, disc; + int error = -EDQUOT; + if (!inode) + return -ENOENT; /* * Find out if this filesystem uses i_blocks. */ @@ -1107,27 +1248,60 @@ case USRQUOTA: if (inode->i_uid == iattr->ia_uid) continue; - transfer_from[cnt] = dqget(inode->i_dev, (direction) ? iattr->ia_uid : inode->i_uid, cnt); - transfer_to[cnt] = dqget(inode->i_dev, (direction) ? inode->i_uid : iattr->ia_uid, cnt); + /* We can get transfer_from from inode, can't we? */ + transfer_from[cnt] = dqget(inode->i_dev, inode->i_uid, cnt); + transfer_to[cnt] = dqget(inode->i_dev, iattr->ia_uid, cnt); break; case GRPQUOTA: if (inode->i_gid == iattr->ia_gid) continue; - transfer_from[cnt] = dqget(inode->i_dev, (direction) ? iattr->ia_gid : inode->i_gid, cnt); - transfer_to[cnt] = dqget(inode->i_dev, (direction) ? inode->i_gid : iattr->ia_gid, cnt); + transfer_from[cnt] = dqget(inode->i_dev, inode->i_gid, cnt); + transfer_to[cnt] = dqget(inode->i_dev, iattr->ia_gid, cnt); break; } - if (check_idq(transfer_to[cnt], cnt, 1, initiator, tty) == NO_QUOTA || - check_bdq(transfer_to[cnt], cnt, blocks, initiator, tty, 0) == NO_QUOTA) { - for (disc = 0; disc <= cnt; disc++) { - dqput(transfer_from[disc]); - dqput(transfer_to[disc]); + /* Something bad (eg. quotaoff) happened while we were sleeping? */ + if (transfer_from[cnt] == NODQUOT || transfer_to[cnt] == NODQUOT) + { + if (transfer_from[cnt] != NODQUOT) { + dqput(transfer_from[cnt]); + transfer_from[cnt] = NODQUOT; + } + if (transfer_to[cnt] != NODQUOT) { + dqput(transfer_to[cnt]); + transfer_to[cnt] = NODQUOT; } - return(NO_QUOTA); + continue; + } + /* + * We have to lock the quotas to prevent races... + */ + if (transfer_from[cnt] < transfer_to[cnt]) + { + lock_dquot(transfer_from[cnt]); + lock_dquot(transfer_to[cnt]); + } + else + { + lock_dquot(transfer_to[cnt]); + lock_dquot(transfer_from[cnt]); + } + + /* + * The entries might got invalidated while locking. The second + * dqget() could block and so the first structure might got + * invalidated or locked... + */ + if (!transfer_to[cnt]->dq_mnt || !transfer_from[cnt]->dq_mnt || + check_idq(transfer_to[cnt], cnt, 1, initiator, tty) == NO_QUOTA || + check_bdq(transfer_to[cnt], cnt, blocks, initiator, tty, 0) == NO_QUOTA) { + cnt++; + goto put_all; } } + if ((error = notify_change(dentry, iattr))) + goto put_all; /* * Finally perform the needed transfer from transfer_from to transfer_to, * and release any pointers to dquots not needed anymore. @@ -1139,28 +1313,39 @@ if (transfer_from[cnt] == NODQUOT && transfer_to[cnt] == NODQUOT) continue; - if (transfer_from[cnt] != NODQUOT) { - dquot_decr_inodes(transfer_from[cnt], 1); - dquot_decr_blocks(transfer_from[cnt], blocks); - } + dquot_decr_inodes(transfer_from[cnt], 1); + dquot_decr_blocks(transfer_from[cnt], blocks); - if (transfer_to[cnt] != NODQUOT) { - dquot_incr_inodes(transfer_to[cnt], 1); - dquot_incr_blocks(transfer_to[cnt], blocks); - } + dquot_incr_inodes(transfer_to[cnt], 1); + dquot_incr_blocks(transfer_to[cnt], blocks); + unlock_dquot(transfer_from[cnt]); + dqput(transfer_from[cnt]); if (inode->i_dquot[cnt] != NODQUOT) { struct dquot *temp = inode->i_dquot[cnt]; inode->i_dquot[cnt] = transfer_to[cnt]; + unlock_dquot(transfer_to[cnt]); dqput(temp); - dqput(transfer_from[cnt]); } else { - dqput(transfer_from[cnt]); + unlock_dquot(transfer_to[cnt]); dqput(transfer_to[cnt]); } } - return(QUOTA_OK); + return 0; +put_all: + for (disc = 0; disc < cnt; disc++) { + /* There should be none or both pointers set but... */ + if (transfer_to[disc] != NODQUOT) { + unlock_dquot(transfer_to[disc]); + dqput(transfer_to[disc]); + } + if (transfer_from[disc] != NODQUOT) { + unlock_dquot(transfer_from[disc]); + dqput(transfer_from[disc]); + } + } + return error; } @@ -1224,16 +1409,18 @@ struct vfsmount *vfsmnt; struct file *filp; short cnt; + int enabled = 0; + /* We don't need to search for vfsmnt each time - umount has to wait for us */ + vfsmnt = lookup_vfsmnt(dev); + if (!vfsmnt || !vfsmnt->mnt_sb) + goto out; + + /* We need to serialize quota_off() for device */ + down(&vfsmnt->mnt_dquot.dqoff_sem); for (cnt = 0; cnt < MAXQUOTAS; cnt++) { if (type != -1 && cnt != type) continue; - - vfsmnt = lookup_vfsmnt(dev); - if (!vfsmnt) - goto out; - if (!vfsmnt->mnt_sb) - goto out; if (!is_enabled(vfsmnt, cnt)) continue; reset_enable_flags(vfsmnt, cnt); @@ -1242,26 +1429,25 @@ reset_dquot_ptrs(dev, cnt); invalidate_dquots(dev, cnt); + /* Wait for any pending IO - remove me as soon as invalidate is more polite */ + down(&vfsmnt->mnt_dquot.dqio_sem); filp = vfsmnt->mnt_dquot.files[cnt]; vfsmnt->mnt_dquot.files[cnt] = (struct file *)NULL; vfsmnt->mnt_dquot.inode_expire[cnt] = 0; vfsmnt->mnt_dquot.block_expire[cnt] = 0; + up(&vfsmnt->mnt_dquot.dqio_sem); fput(filp); - } + } /* * Check whether any quota is still enabled, * and if not clear the dq_op pointer. */ - vfsmnt = lookup_vfsmnt(dev); - if (vfsmnt && vfsmnt->mnt_sb) { - int enabled = 0; - for (cnt = 0; cnt < MAXQUOTAS; cnt++) - enabled |= is_enabled(vfsmnt, cnt); - if (!enabled) - vfsmnt->mnt_sb->dq_op = NULL; - } - + for (cnt = 0; cnt < MAXQUOTAS; cnt++) + enabled |= is_enabled(vfsmnt, cnt); + if (!enabled) + vfsmnt->mnt_sb->dq_op = NULL; + up(&vfsmnt->mnt_dquot.dqoff_sem); out: return(0); } @@ -1282,47 +1468,51 @@ if (is_enabled(vfsmnt, type)) return -EBUSY; - mnt_dquot = &vfsmnt->mnt_dquot; + mnt_dquot = &vfsmnt->mnt_dquot; + down(&mnt_dquot->dqoff_sem); tmp = getname(path); error = PTR_ERR(tmp); if (IS_ERR(tmp)) - return error; + goto out_lock; f = filp_open(tmp, O_RDWR, 0600); putname(tmp); - if (IS_ERR(f)) - return PTR_ERR(f); - /* sanity checks */ + error = PTR_ERR(f); + if (IS_ERR(f)) + goto out_lock; error = -EIO; if (!f->f_op->read && !f->f_op->write) - goto cleanup; + goto out_f; inode = f->f_dentry->d_inode; error = -EACCES; if (!S_ISREG(inode->i_mode)) - goto cleanup; + goto out_f; error = -EINVAL; if (inode->i_size == 0 || (inode->i_size % sizeof(struct dqblk)) != 0) - goto cleanup; + goto out_f; - /* OK, there we go */ set_enable_flags(vfsmnt, type); mnt_dquot->files[type] = f; dquot = dqget(dev, 0, type); - mnt_dquot->inode_expire[type] = (dquot) ? dquot->dq_itime : MAX_IQ_TIME; - mnt_dquot->block_expire[type] = (dquot) ? dquot->dq_btime : MAX_DQ_TIME; + mnt_dquot->inode_expire[type] = (dquot != NODQUOT) ? dquot->dq_itime : MAX_IQ_TIME; + mnt_dquot->block_expire[type] = (dquot != NODQUOT) ? dquot->dq_btime : MAX_DQ_TIME; dqput(dquot); vfsmnt->mnt_sb->dq_op = &dquot_operations; add_dquot_ref(dev, type); - return(0); + up(&mnt_dquot->dqoff_sem); + return 0; -cleanup: - fput(f); - return error; +out_f: + filp_close(f, NULL); +out_lock: + up(&mnt_dquot->dqoff_sem); + + return error; } /* @@ -1343,6 +1533,9 @@ if ((u_int) type >= MAXQUOTAS) goto out; + if (id & ~0xFFFF) + goto out; + ret = -EPERM; switch (cmds) { case Q_SYNC: @@ -1350,7 +1543,7 @@ break; case Q_GETQUOTA: if (((type == USRQUOTA && current->euid != id) || - (type == GRPQUOTA && current->egid != id)) && + (type == GRPQUOTA && in_group_p(id))) && !capable(CAP_SYS_RESOURCE)) goto out; break; @@ -1360,7 +1553,7 @@ } ret = -EINVAL; - dev = 0; + dev = NODEV; if (special != NULL || (cmds != Q_SYNC && cmds != Q_GETSTATS)) { mode_t mode; struct dentry * dentry; --- linux/fs/super.c Wed Jun 16 22:08:10 1999 +++ linux/fs/super.c Wed Jun 16 22:25:14 1999 @@ -104,7 +104,8 @@ lptr->mnt_dev = sb->s_dev; lptr->mnt_flags = sb->s_flags; - sema_init(&lptr->mnt_dquot.semaphore, 1); + sema_init(&lptr->mnt_dquot.dqio_sem, 1); + sema_init(&lptr->mnt_dquot.dqoff_sem, 1); lptr->mnt_dquot.flags = 0; /* N.B. Is it really OK to have a vfsmount without names? */ --- linux/fs/inode.c Wed Jun 16 21:55:30 1999 +++ linux/fs/inode.c Thu May 13 22:30:38 1999 @@ -44,7 +44,7 @@ * allowing for low-overhead inode sync() operations. */ -static LIST_HEAD(inode_in_use); +LIST_HEAD(inode_in_use); static LIST_HEAD(inode_unused); static struct list_head inode_hashtable[HASH_SIZE]; --- linux/include/linux/quotaops.h Wed Jun 16 21:59:26 1999 +++ linux/include/linux/quotaops.h Thu May 13 22:31:12 1999 @@ -31,8 +31,8 @@ extern void dquot_free_block(const struct inode *inode, unsigned long number); extern void dquot_free_inode(const struct inode *inode, unsigned long number); -extern int dquot_transfer(struct inode *inode, struct iattr *iattr, - char direction, uid_t initiator); +extern int dquot_transfer(struct dentry *dentry, struct iattr *iattr, + uid_t initiator); /* * Operations supported for diskquotas. @@ -101,15 +101,10 @@ if (dentry->d_inode->i_sb->dq_op) { if (IS_QUOTAINIT(dentry->d_inode) == 0) dentry->d_inode->i_sb->dq_op->initialize(dentry->d_inode, -1); - if (dentry->d_inode->i_sb->dq_op->transfer(dentry->d_inode, iattr, 0, current->fsuid)) - goto out; - error = notify_change(dentry, iattr); - if (error) - dentry->d_inode->i_sb->dq_op->transfer(dentry->d_inode, iattr, 1, current->fsuid); + error = dentry->d_inode->i_sb->dq_op->transfer(dentry, iattr, current->fsuid); } else { error = notify_change(dentry, iattr); } -out: return error; } --- linux/include/linux/fs.h Wed Jun 16 21:59:00 1999 +++ linux/include/linux/fs.h Thu May 13 22:31:12 1999 @@ -639,7 +639,7 @@ int (*alloc_inode) (const struct inode *, unsigned long, uid_t); void (*free_block) (const struct inode *, unsigned long); void (*free_inode) (const struct inode *, unsigned long); - int (*transfer) (struct inode *, struct iattr *, char, uid_t); + int (*transfer) (struct dentry *, struct iattr *, uid_t); }; struct file_system_type { --- linux/include/linux/mount.h Wed Jun 16 21:57:32 1999 +++ linux/include/linux/mount.h Mon May 24 22:55:13 1999 @@ -17,11 +17,12 @@ struct quota_mount_options { unsigned int flags; /* Flags for diskquotas on this device */ - struct semaphore semaphore; /* lock device while I/O in progress */ + struct semaphore dqio_sem; /* lock device while I/O in progress */ + struct semaphore dqoff_sem; /* serialize quota_off() and quota_on() on device */ struct file *files[MAXQUOTAS]; /* fp's to quotafiles */ time_t inode_expire[MAXQUOTAS]; /* expiretime for inode-quota */ time_t block_expire[MAXQUOTAS]; /* expiretime for block-quota */ - char rsquash[MAXQUOTAS]; /* for quotas threath root as any other user */ + char rsquash[MAXQUOTAS]; /* for quotas threat root as any other user */ }; struct vfsmount | |